AI media analysis Archives - The Media Copilot

The future of journalism is personal: How The Journal is building AI for readers, not robots

The Copilot — Thu, 25 Jun 2026 13:03:10 +0000

This episode is sponsored by: Adobe Acrobat

This week on The Media Copilot, Pete Pachal sits down with Taneth Evans, Head of Digital at The Wall Street Journal, to explore how one of the world’s leading news organizations is navigating the AI revolution.

Rather than chasing every new AI trend, Evans shares how the Journal evaluates emerging technology through a simple lens: Does it genuinely help journalists do better work or help readers better understand the world?

From AI-powered investigative tools and newsroom workflows to personalized storytelling and adaptive content, Evans offers a thoughtful look at how AI can strengthen journalism without compromising trust.

“So many times in the past few years, I’ve said to people, what would you do with a building full of journalists at your disposal? No newsroom feels like it has enough resources… How can we use AI to help us get closer to the answers to that question?” — Taneth Evans

The conversation explores why journalism is evolving beyond a single article format into flexible experiences tailored to how each reader prefers to consume information, while keeping facts, reporting, and editorial standards at the center.

The new Adobe productivity agent orchestrates tools and models to generate images, text and rich content like presentations, podcasts and social posts, while also powering conversational PDF editing in Acrobat.

With new PDF Spaces capabilities, users can combine files, links and notes into interactive, shareable spaces for research, collaboration and content creation. VICE News, Kid Cudi and celebrity event planner Mindy Weiss are already using these tools to build trust and deeper engagement with their audiences.

Link: Do that with Acrobat: AI-Powered PDF workspaces | Adobe Acrobat

What we cover

• How The Wall Street Journal evaluates new AI technologies

• Why audience needs come before AI innovation

• The rise of personalized and adaptive journalism

• AI tools transforming investigations and newsroom workflows

• How AI can create entirely new reader experiences

• Why trust, attribution, and media literacy matter more than ever

• The future of publisher owned experiences in an AI driven world

• Why great reporting becomes even more valuable in the age of AI

As AI changes how information is distributed, the challenge isn’t simply adopting new technology. It’s preserving trust while creating better ways for people to engage with journalism. Evans argues that the future belongs to news organizations that use AI to deepen their relationship with readers, not replace it.

Why this matters

News consumption is changing rapidly. Readers increasingly expect personalized, accessible experiences while publishers face growing competition from AI powered search, chatbots, and automated summaries. The organizations that succeed will be those that combine trusted reporting with innovative experiences that make journalism more useful, more engaging, and more relevant. This conversation offers an inside look at how one of the world’s leading newsrooms is preparing for that future.

About the 👤 Guest

Taneth Evans Head of Digital, The Wall Street Journal

LinkedIn: https://www.linkedin.com/in/taneth-evans-b35877162/

The Wall Street Journal: https://www.wsj.com

About the show: To explore more conversations like this and see what’s new, visit the Media Copilot website at mediacopilot.ai. You’ll find new episodes, expanded resources, and tools designed for journalists, communicators, and media leaders navigating the fast-changing world of AI. It’s the home base for everything Media Copilot and it’s just getting started.

Enjoyed this episode?

Subscribe to The Media Copilot on Substack, Apple Podcasts, Spotify, or your favorite app. On YouTube? Tap the Like button and Subscribe to the YouTube channel. For more AI tools and resources built for media professionals, visit mediacopilot.ai.

Produced by Pete Pachal and Executive Producer Michele Musso
Edited by the Musso Media Team

Music: “Favorite” by Alexander Nakarada, licensed under CC BY 4.0

TRANSCRIPT

Pete Pachal (00:25.442)

Hi, welcome to the Media Copilot. It’s a podcast about how AI is changing media, news, and communication. My name’s Pete Pachael. I cover tech for a long time as a journalist, and now I have deep conversations with the media people, the builders, and the creators, or are all answering the question how will we get information in the future? And how will that change the jobs in the industries whose business is information, especially media?

One of the biggest questions in media right now is whether AI will make journalism more useful or more generic. We already live in a world of apps and services that are trying to make information as convenient and efficient as possible. I think AI summaries, chatbots, personalized feeds, the list goes on. That creates a huge challenge for publishers, especially premium publishers. AI might be able to help you serve your readers better, but how do you do that without flattening the reporting or weakening the brand or worst of all, breaking trust?

My guest this week sits right in the middle of that question. Taneth Evans is head of digital at the Wall Street Journal, which is obviously one of the most important news organizations in the world. Her work touches everything from audience strategy to newsroom culture to the product roadmap, and that includes the journal’s approach to AI. The journal’s already moved forward with some AI-driven features, including AI summarized bullet points, some reporter tools, and even a bespoke chatbot that was specifically made for iPhone coverage.

But what I think makes the journal’s approach interesting isn’t just the tools. It’s the broader idea that journalism itself may become more flexible. The same reporting can turn into different formats, whether that’s summaries, explainers, audio, video, interactive experiences, or even something else, all depending on what the reader wants. So today we’re going to talk about how the journal thinks about AI, how a global newsroom with serious standards decides what’s safe to ship.

And what audience ac audiences actually want from AI-powered news products. I’m excited to get into it. quick note: if you’re listening on Apple or Spotify, please leave a five-star review and maybe a nice comment. And if you’re watching on YouTube, please like the video and subscribe to the channel if you don’t mind. Those things really do help more people find the show. All right, let’s get started. Taneth Evans, welcome to the Media Copilot.

Taneth (02:40.285)

Thanks for having me.

Pete Pachal (02:42.35)

it’s my pleasure. so I want to get into all that cool stuff, AI experiences, what everything you are in charge of there at the journal and and how it’s rapidly progressing along with AI. But I’d love to hear a little bit more about you and your background. So tell me a little bit about, you know, how you how you came to be at the journal, how your role has evolved there, and if in particular like how it’s evolved alongside AI.

Taneth (03:07.345)

Hmm. I’ve been at the journal for just over three years, three years in February. I arrived with the new editor-in-chief, Tucker, new at the time. I had worked with her previously in London, where she was editor of the Sunday Times, and I was lucky enough to be brought along for the ride when she came to the journal. And so it’s actually, although it’s been three years, feels like…

a lot longer because well humbly what we’ve achieved feels like more than three years work. We arrived and kind of set a broad newsroom strategy and spent a long time articulating that and making sure we had all of the resources and skill sets we needed to action it. And then as you say, think we were, know, AI technology was obviously very important three years ago and it was obviously going to become more important.

But I think we were all surprised, I certainly was, by the speed with which it started taking over lots of conversations, both in the workplace and beyond, really. And so it did become a larger part of my job very quickly. It’s interesting when you think about who should own these things in a newsroom. There’s so many things that come along with it. It’s technology, yes, but it’s also…

governance and strategy and to what end we will employ this technology. And so I think I was kind of in a very lucky position really to be the obvious person to have it sit within my team. But we very quickly spun up a working group in the newsroom that was led by our now head of data and AI, the wonderful Tess Jeffers. And I think that’s been one of the really important things

that we created that working group very quickly. Firstly, to talk about governance and guidelines and how we would speak to the newsroom about AI. But now it means that that group is on the forefront of discussing emerging technology and stress testing it and thinking about how it should look and work in the newsroom. And so nothing is handed to us. We’re very much on the front foot of creating those guidelines and

Taneth (05:28.765)

hopefully making it a little bit exciting and less intimidating too.

Pete Pachal (05:34.538)

So you this working group has sort of evolved into the main sort of filter, I guess, as new technologies come out and and new techniques sort of develop to that does it go through this this group and and what is that process like?

Taneth (05:51.504)

Yeah, there are representatives from all corners of the newsroom in that group. there are people from my team, the digital team, content strategy, audience development, all of that good stuff. There are also people who represent video, audio, all of the different formats, visual storytelling on the website. But then we also have representatives from standards and the investigations team and

reporters that are very active and interested in using the technology. And so it means that those discussions are lively and exciting and we really kind of drill into things very deeply. And I think that’s why it’s been successful. It’s full of people that are very curious and excited. And so we can, you know, look at a proposition or a new technology and very quickly say, but practically, what does this mean?

Pete Pachal (06:45.954)

Mm-hmm.

Taneth (06:46.552)

what end would we use this rather than, you know, talking about things.

Pete Pachal (06:51.554)

And how’s like how what is your filter when you approach that, right? Because you can think about it on across a number of dimensions. You can think about it like, you know, product and efficiencies. You can think about, you know, what the audience is it enhancing some kind of experience or just getting things to them quicker? you know, overall distribution, I’m sure that sort of factors in. and are there any like how how do you like to approach these and you know what within those filters, like

How do you sort of identify red lines and sort of like how where you would s definitely say no?

Taneth (07:24.731)

It’s interesting, isn’t it? Because people will often say, so what are you doing with AI? And it’s kind of, it’s too big of a question. What does, what does it mean? Because as you rightly said, it’s, it’s so many things. It’s a distribution layer, one that I think will fundamentally change, maybe has already fundamentally changed how consumers will act out there in the wild. It’s internal tooling, yes, efficiencies, but also

things to make us more powerful. It’s experience. It’s, know, ways that we might augment our products and offer net new things to readers. And so the first, for me, the first layer we have to go through is, this a need? Does this meet a need? Does it meet a need that we have in the newsroom or does it meet, crucially, one of our audience needs? Because I don’t, I’m not so interested in doing things that

will be cool and that no one will look at. I love doing cool things, don’t get me wrong, but I want people to see it. I want it to be for a reason, you know? And the other, I suppose big thing that I try to advocate for is in the newsroom, I want us to think about using AI for net new, for powerful things. I already alluded to it, but efficiencies are great and we all need them, but I’m…

less interested in them. I’m really, really interested in the stuff that we can do that we wouldn’t be able to do before. So many times in the past few years, I’ve said to people, what would you do with a building full of journalists at your disposal? No newsroom feels like it has enough resources. I mean, ever, but particularly now. How can we, how would we answer that question? And how can we use AI to help us get closer to the answers to that question?

Pete Pachal (09:06.722)

Right.

Pete Pachal (09:18.146)

Nice. So what what would you say is been net new? Like what what has excited you the most over the past couple of years and where do you see that particular aspect, not just the efficiencies, but the the new stuff? Where’s that going?

Taneth (09:32.333)

In terms of stuff that we’ve already shipped or started using, we have a lot of internal tooling that I think is really exciting and that I’m excited about. Internal tooling doesn’t sound exciting, does it?

Pete Pachal (09:43.534)

Believe me, I’m excited about it. My reader my listeners are excited about it. We’d love to hear about the internal tool bit.

Taneth (09:49.32)

Good. Well, firstly, in investigations, I think that’s the very obvious place that lots of people have started. It’s allowing us to parse and pattern match within kind of large swathes of information in a way that maybe humans could do, maybe in some places they couldn’t. Similarly, helping us to pinpoint information in large documents or large pieces of information. Again, humans…

could do that, it would take us a lot, lot longer without this technology. And then similarly, we have built within the newsroom, a proprietary tool that we call Orca. That is a tool that turns messy audio into structured data so that we can do lots more with audio files. for example, we took

over 2200 hours of podcasts and had Orca listen to them and help us search around information in order to write a report on how the MAGA base and particularly the podcasting coming from that were reacting to the Epstein files and how that changed their attitude towards the government. That is something that we could have done without this tool but it would have taken

a really long time. was, as I say, huge, huge amounts of audio. And so again, this is something that we just simply could not have turned around in that time without this technology. And so although maybe not net new, I think it’s creating net new outputs in terms of speed and scale of our storytelling, certainly. And then away from investigations within our news wires, our news wires audience is slightly different to the general audience, of course.

and so they have different needs. One thing that we’ve done that I’m really excited about is a new feature called company talks. and that is AI generated reports based on company announcements. So we will take a company press release. We will allow an AI generated report to be created and then an editor will check it. That is net new coverage that we before didn’t offer to our newswires audience. So it’s additive to the experience and similar.

Pete Pachal (12:13.708)

And to be to be clear on that, I just want to clarify that. So it’s like it’s not just rewriting the release, I assume. It’s like it’s you’re bringing context in the journal’s history of report whatever I you tell me, like it’s bringing context to to it, correct?

Taneth (12:27.315)

We’re telling readers what it means. But also similarly, because it’s a news-wise audience, actually often they just want a well-written understanding of what that report says. Exactly. Yeah. We’ve also been able to offer our news wires in different languages using AI translation. So we now have Chinese, Japanese, Korean, French, German. Again, it’s a kind of net new offering that we can take out to clients.

Pete Pachal (12:34.732)

I see. In an expected sort of templated way. I get it. Yep.

Pete Pachal (12:55.01)

Nice. Go on.

Taneth (12:56.487)

Go on. And then one that we haven’t yet shipped, I think is exciting. So we should talk about it is in the new experience kind of realm. So we’re working on a product right now called Backstory, which will live in our app and it will allow readers to on a given article, understand the context and the background to that story. And one reason I’m really excited about this feature is that it’s really come from a need both

Pete Pachal (13:02.616)

Cool.

Taneth (13:26.277)

internally and of our audiences. So I’m sure you’ve experienced this, that when writing on a long running storyline or topic, you have to put in the B matter, you have to put in the background because it could be that a reader’s coming fresh to that story. Sometimes it weighs our stories down and we want to kind of trim them or get to the new stuff more quickly, but our readers have an expectation that we catch them up if they don’t know.

what’s already happened. The backstory will allow us to offer that to the readers that need it, but for the readers that don’t, get them more quickly into the crux of the new information. I’m really excited about that because I think it takes a problem that we have had for a long time as an industry, and it really kind of almost revolutionizes how we might tell those stories.

Pete Pachal (14:18.03)

Yeah, I think most reporters sort of default to somewhat and again, this is no fault of their own. They’re great on their beats, but they’ll default to sort of speaking to people who are sort of keeping up with everything they’re read writing. And as a as a newsroom leader, I and and sort of a long t you I I had to sort of beat that out of myself as I was doing. And so now I find I kind of do the opposite. But to your point, those are different readers and they both deserve to be served, right? Like in other words, don’t bog down the people who are keeping up, but also there are

take into account the people that haven’t necessarily been following every development. And if that can burden can shift from the the writer or even the editor and AI can sort of take on some of this interpretability, that that becomes very interesting and potentially powerful. And that’s kind of what I wanted to steer toward because I know you’ve written about this, about adaptive content and sort of the the next

sort of phase of that to me feels like what just generally called liquid content, which is like, you have all this facts and reporting and and you can turn that into whatever, you know, you could turn it into an article, you can turn it into podcast, etc. So, sort of break down your thinking on that, on sort of where that can live, because we’ve been talking about sort of newsroom tools. I’m sort of shifting that a little bit into a reader experience. how do you how do you sort of organize your thoughts around like what

the tool is a tool for the storytellers and what’s a tool for the people interpreting it.

Taneth (15:48.964)

Mm-hmm. So I’m very excited about this. And I have been thinking about this for a long time. And when I first started thinking about what people now refer to as liquid content, I was like, I’m a genius. I’ve cracked it. I fixed journalism. And then I became obsessed with it. And I was talking about it and quickly realized that, of course, I was not the only person to see this technology and have this idea. And that’s because it’s a really real problem, I think. that is that personalization, I think,

Pete Pachal (16:02.988)

Ha ha ha.

Taneth (16:18.195)

We have not cracked personalization as an industry. New generations of readers expect things to be highly personalized, whether they explicitly tell us that, or whether just in their everyday experiences on social media, on shopping channels, on Netflix, they are used to seeing things that they like and that they want to engage with. At the same time, we all lived through

personalisation of social platforms doing not great things for news. You know, those feeds became highly personalised and people ended up in quite concerning, sometimes filter bubbles. And so as an industry, we really didn’t want to exacerbate that problem. And so we have, think, shied away from personalisation. Not that’s a kind of real overgeneralisation. And of course, lots of news publishers are doing

really cool things with personalization. But I think as an industry, you know, we’re not offering yet highly personalized experiences. And I think that’s partly because we’ve only looked at personalization through the lens of topic. And we’re saying, this is a really important story, and we don’t want to hide it from people. And actually, it’s the same from a reader perspective. For us at the journal, readers tell us that they want a curated experience, they want us to tell them

what we are seeing as the most important stories that day. so personalization of topic becomes a little sticky for all of those reasons. I think this technology has allowed us all to think about personalization in a slightly different way. And that’s in personalization of format of how a reader might consume something. And that could look like lots of different things. It could look like, me this story, but give it to me.

in audio or in video you know I’m on my commute and it’s 13 minutes long give me a 13 minute audio version of this story it could be that we see that Pete only ever watches videos on the journal app give it to him in a video version I think that’s the first step I think we take it further by personalizing the actual story itself so not just format but

Taneth (18:38.373)

If we see that I engage better with stories that are led with a case study because I’m an empath and I need to see how it affects humans, then what if our building blocks of that story could be rearranged in such a way that we give me that version? And if the story is about a company that I invest in, so I actually just want the numbers really quickly up top, let’s give that to me instead. And so then your information becomes kind of

building blocks that could be arranged on the fly for the right person in a way that is the right way to consume for them.

Pete Pachal (19:17.218)

Hmm. Yeah, it seems like as you were speaking about that there, that the way the the way I am understanding how memory is working in these models, you know, they sort of build up this file over time. and I know OpenAI is doing sort of even more advanced things, but it does feel like this is the media equivalent of that, that it’s like it’s it’s sort of taking that idea of memory. I was like, and it so it’s not even anything I I I specify.

in the app, it just sort of understands, this is from my behavior. I’m doing this. So starts to build up this memory. And sure, there might be some setting I just turn on at the very beginning of this, but over time the experience of coming to the journal would just evolve to just match my needs and not at both sort of the app level, the story level. is that sort of a fairly accurate picture of kind of what you’re you’re thinking about?

Taneth (20:11.405)

Yeah, and I do think audiences are going to expect it. They are going to come to expect it. Especially as, you know, time is our biggest competitor and younger generations are turning away from the news. And one of the reasons we know from studies that one of the reasons that they’re turning away is that they don’t feel that the news is relevant to them. And I think this goes a long way into taking

the news of the day and giving it to people in a relevant way that they feel will impact their lives.

Pete Pachal (20:46.476)

Right. I know you didn’t mean time the publication there, but I was just to clarify for the listeners, I was like, Wait, time is? No.

Taneth (20:51.315)

No, time the concept, I’m sorry.

Pete Pachal (20:58.252)

Yeah. Yeah, no, I get it. I get it. so interesting. So we’re talking about liquid content adapting this stuff. It’s great. So what what changes about the journalist’s job then as a reporter or editor? Anything? do they have to sort of get ahead of some of these formats? and if it sort of becomes this thing where the final content’s malleable, I think sometimes reporters fear that, well, I’m just like a fact.

Taneth (21:01.171)

you

Pete Pachal (21:24.588)

I’m putting facts in a in a robot or an engine and it’s just creating things with it. tell me tell me what your vision is on the sort of news production side.

Taneth (21:34.558)

think firstly, there will always be a place for a well-written narrative yarn. know, we have seen books are still here. People have predicted the end of books for a long time and they’re still here because we want them. And I think similarly, kind of long, well-reported narrative pieces of journalism.

will survive. That’s the first thing that I should say. But there are some forms of journalism that are there to deliver new pieces of information. And I think that is where this kind of technology will play. And so I do see a future that a really brilliant reporter will spend the majority of their time going out and getting facts and filing them in those building blocks.

the new piece of information, the quotes, the characters, the rights of reply, the images, all of these kinds of pieces of metadata that we can use technology to build many, different end results. And look, frankly, I think a lot of reporters will be pleased to hear that. Lots of reporters tell me that the best part of their job is going out and finding the facts. And so I think reporting is going to become

a very premium requirement, you know, like I think it’s going to be more important than ever. And but that’s not to say that there won’t be it’s kind of a spectrum, you know, that everything’s everything’s a spectrum, there will be I really do believe then a premium on the also the really in depth well written kinds of journalism to

Pete Pachal (23:25.708)

Nice. And you you mentioned earlier, you know, there’s some reporters who sit on the committee you talked about, and they’re all obviously probably enthusiasts. I’m sure within the newsroom there’s a spectrum there too of people who are all in on this. It’s great. They see it as a very great tool, and some folks that might need some some coaxing. And can you give me a sense of kind of what the transition’s been like? I mean, transition, I guess, in terms of like the AI era, cause some come to think of it, but like

How have you been able to get catch up some of those folks who might be a little skeptical of like this whole AI this all AI thing and how it affects their job and their industry?

Taneth (24:07.513)

Mm-hmm. I think that it’s actually been an interesting challenge because it has come at us with such a speed, this technology, that everyone, you could take a room of people and everyone would have a slightly different level of experience or understanding. So other things you can kind of launch training sessions and get everyone together and talk about it. This is, it has been quite a unique challenge.

And one way that we’ve tackled it in the newsroom is by running kind of brown bag, know, lunch and lunch, come along and see how other people are using this technology in their work. They’ve been really successful sessions because I think AI can sometimes feel like we’re doing a lot of like kind of broad talking about it. And then you kind of get to it and you’re like, well, I kind of had that myself. I was like, yeah, AI, great. And then I sat down and I was like, hmm.

So what should I do? And it was kind of only when I started to practically get my hands dirty that I was like, starting to see more opportunities within my kind of personal sphere and workflow. And so those brown bags have been a great learning tool for the newsroom because they’ve seen how their colleagues are actually practically using things and that sparked ideas. It also means that we can have good like no dumb questions sessions that people can.

really at any level come and say help me. Again, all credit for this must go to my head of data and AI Tess who’s run these sessions. And the next thing that she’s running during the summer are a series of vibe coding sessions. So again, getting people in, it’s this thing that they all kind of vaguely know about and talk about, but then practically don’t know where to start. And so again, it’s kind of putting the technology into everyone’s hands and saying,

Okay, come on, let’s talk about your ideas and where you might use it.

Pete Pachal (26:05.516)

Nice. Yeah, these vibe coding I think is obviously very powerful, but it also has this thing that it it if it’s not managed well, it feels like you know, it just becomes this wild west and people kind of doing duplicative stuff sometimes. I’m not sure what stage you’re at or whatever, what you’re thinking about, if there’s a long term vision for that. But I’m curious if you’re thinking about like how you would transition from someone doing something interesting and very cool with vibe coding and

putting that into like some actual product if there’s enough innovation there and enough interest.

Taneth (26:40.023)

Mm We’ve seen a few examples of internal tooling. So Brian, who’s a member of our social media team, vibe coded a solution for creating social posts, which I just totally oversimplified. Poor Brian. And he kind of brought it to the AI working group who took a look at it said, Yeah, pretty cool. And then we were able to

Pete Pachal (26:56.195)

Mm-hmm.

Taneth (27:07.703)

ship it and offer it to the whole social media team. We’re lucky that we have good allies in product who can help us kind of jump in and make these things a reality. But I think we’re still kind of starting with that. But again, I must give all credit to the working group. They are the front line of this, you know, the ideas come in, they have a look, they sometimes augment them. And they’re really very aware of everything that’s happening in the newsroom. To go back to your point of kind of

launching similar things and everyone kind of having similar problems. think again, we don’t want to innovation. You want people to get their hands dirty and try things, but then equally you don’t want 10 different tools out in the world doing the same thing. So I think right now our work is at such a scale that it can go through TAS and the working group. And so that really helps us. What that looks like.

when we have a thousand people vibe coding things, TBC.

Pete Pachal (28:13.036)

Nice. So I know you guys did some stuff with chat bots and chat experiences. I had Joanna Stern on back when you had first launched the Joanna bot a bit ago. I she’s no longer with the journal, but the I was curious what you learned from that specifically and and what your take is on like chat bots in general vis a vis media sites. Do you have a do you have any thoughts there?

Taneth (28:36.369)

Yeah, I’m reluctant to launch a chatbot, capital A, capital C. I think not just with AI, we’ve seen for many, many years that if you put too much expectation onto a reader or a user, they become overwhelmed and they don’t know where to start. I mean, of course they do. If you said to someone, talk to this generic chatbot about the Wall Street Journal.

Pete Pachal (28:40.386)

Hmm.

Taneth (29:02.503)

you know, where would they start? And it’s the same with when we make, you know, user interaction, but when we build user interaction into our journalism, we have to prompt, we have to help them. It’s, you know, that’s on us. And so I think the Joanna Bot works so well because it was a specific kind of niche thing. Readers were prompted, they were helped along. We also did the same with Lars, which was our tax bot, which it was utility.

offered readers a service and said this is what we will give you. You can ask us your very specific questions when it comes to tax for example. I think that’s what you need to do. You need to help someone into the experience because if you offer them look at anything wherever you want, whenever you want, however you might like to do it, I think we’d be disappointed in the engagement rate.

Pete Pachal (29:32.334)

Mm-hmm.

Pete Pachal (29:55.021)

Yeah, I I totally I’m seeing that too. In other words, like the more successful ventures into this idea of a chat experience are always sort of super targeted, whether it’s something like on iPhones or taxes or in other places I’ve seen election stuff. you know, we’ll see how it evolves. But I also think this transitions nicely into the whole idea of discovery because I think that also like it’s like what do you expect from a media

specific experience on something like the journal versus like your broad chat GPT perplexity Google AI overview experience, right? And so like you know, I sometimes feel like media companies doing chatbots is like media companies doing Facebook or you know, their own social network or something like that. It’s just it’s I don’t not what we’re looking for. but again like it’s all to me it’s it that transition, okay, well what are they looking for from things off the journal? And how does the journal content then interact with that?

and so obviously more and more people are using AI to to use information. the journal is pretty famously in some some has deals and and lawsuits among the major AI companies. Won’t get into that. But I’m I am just curious on how you in your role think about the the broader public getting good information from these and the journal sort of being a part of that in some

Taneth (31:17.244)

Yeah, it’s interesting, isn’t it? Because there are two like broad routes you could go. There’s one to

kind of go all in on your own product and try to get people directly. And there’s one to make your journalism as easily possible as possible so that people can still encounter it. And I don’t think that’s an AI specific problem. I think it’s been the same. We had Facebook and some articles. We’ve had all of these. These questions have arisen before. I think broadly, we don’t want to disappear.

Pete Pachal (31:29.39)

Mm-hmm.

Taneth (31:54.897)

is a good example. is a slight aside from AI, but TikTok is a great example. I’m not driving traffic from TikTok, but I want us to be on there because I want that generation of users to know that we exist. And it’s a similar thing. So, you know, we are thinking about how we show up in those experiences and making sure that our information is visible. But really what this has doubled down for me is the idea of us being a destination in ourselves. I think

in, I don’t know, 10 years time, maybe, maybe I’m overshooting that even. think websites won’t be visited by humans, they’ll be visited by our agents who are coming to collect the information of the day. But our apps will become really important because that will be the human touch point. It will be people coming directly to the journal. And so we need to give them very, very good reasons to have that direct relationship with us to come.

directly to us and not get our information elsewhere. And so that means our information being excellently accessible, yes, and in a wonderful, pleasing experience. But it also means offering other things. Community, you know, how might you come and talk to other readers about our journalism? Live events, coming and seeing us in person, Chakra, having actually a human relationship with us. It could be other…

features that you can only get at the journal. We need to think really, really carefully about fostering that direct relationship with readers at the same time as allowing our information, I think, to appear in a well attributed way in other experiences.

Pete Pachal (33:38.467)

Yeah, and and Fr I take it then you practice some amount of you know generative engine optimization in what you’re doing to make sure it’s machine readable and that you know I know there’s that’s there’s a difference between that and like just blocking unauthorized crawling, right? But if it is authorized crawling, you know, like you want to make it as as as machine friendly as possible. Is that fair to say that that’s the approach?

Taneth (34:02.076)

Yeah, yeah, but I mean, and I say it kind of like, what is that optimization? You know, like, we’re all kind of talking about it as if it’s a thing. And I’m like, is it a thing? I don’t know. I mean, I’m seeing a lot of promises that you can optimize in different ways. And I think there are things that we can do to make sure that our, you know, our journalists information is up to date and that we are very, you know, our metadata is good, but equally.

I don’t think there are like tricks yet. Maybe there will be, but I think, and actually I think the same about SEO too, that I think build a good website in a way that the internet works and do really good journalism and good things will follow. mean, you sure, I’m sure that’s a purist view, I, I, I’m skeptical that you can really kind of game this.

Pete Pachal (34:34.904)

Right.

Pete Pachal (34:53.794)

No, no.

Pete Pachal (34:58.552)

Well, I also think AI is evolving so rapidly that what rules you might sort of conclude at now might might be radically different as they get sort of better at interpretability. That said, I’m happy to do a brown dagger anytime you want on what I’ve what I’ve put together on this. I I think about it quite a bit, GEO and that sort of thing. so I I I’ve again I’ve read some of what you’ve written. I really liked what you’ve said before about pulling back from traffic chasing.

And you know, having sort of different KPIs than sort of the traditional ones that I think, you know, everyone’s kind of moving away from because obviously search and social just aren’t aren’t really are are being reduced in terms of their discoverability for content. Can you talk a little bit about like how you’re thinking about success with regard to both, you know, the story level and also just sort of just broadly as the journal as a as an enterprise?

Taneth (35:57.341)

I actually don’t think it has drastically changed for us. We arrived three years ago and quickly articulated to the newsroom that we wanted to become a truly audience first publication. What does that mean? It sounds very straightforward. It sounds simple almost. But if you really stop and think about it, it’s…

quite revolutionary, it’s stopping at every decision and saying, what does the audience need from this? Because often the audience need is not what our instinct says journalists might be.

So the newsroom has really successfully kind of come aboard with this ethos. It means that our journalism is, we have something called the digital pause. We ask everyone to pause at the start of any process to ask who is the audience for this? What do they want? What are their needs? Therefore, what should we do? And I think that’s, and I don’t just think I can see,

in our engagement rates that it’s reactive readers are reacting well to that. They’re spending more time with us. They’re finding our journalism more readily. And they’re canceling their subscriptions at lower rates, which is, I suppose the ultimate goal is our net number of subscribers and the amount of money that we get from them.

Pete Pachal (37:18.146)

Nice.

Taneth (37:26.74)

So in the newsroom, I think we will continue on that path regardless of AI, of creating journalism that people want to read and then once they start reading it, they stick around. I use read and I shouldn’t use read. also, they may be watching it, they may be listening to it, they may be experiencing it in other ways. And so I think…

For me, the hallmark of a good strategy is that you’re not changing it all the time. And so really the goals that we set out three years ago are still our goals now. We’re using different tactics. But ultimately it is to grow our audience and retain them.

Pete Pachal (38:06.636)

Nice. So as we wrap up here, I try to pin down the people I talk to about things they are both worried about and hopeful about for AI and how it’s changing the media ecosystem. So I’d love it if you could give me one of each. What are you what are you what are you kind of losing sleep over with regard to AI? And then what is like the thing that was wow, this would be amazing if if it were to come to fruition.

Taneth (38:34.93)

I’m worried about facts and their attribution. I’m worried about facts or misinformation taking on a life of their own if people are no longer going directly to the source. I am very excited about and supportive of…

AI technology, but I, like other people, can’t help but notice the confidence with which it gives me information. And I fear that if, if we don’t work really hard on media literacy and people questioning facts when they’re not coming from a trustworthy source, then I fear, I fear misinformation.

and similar kind of filter bubbles. I think that’s my kind of existential bit.

Pete Pachal (39:33.55)

No, it’s good. Hallucinations are I do feel like they’re kind of inherent to the technology, in my experience anyway. Every time there’s a good a new model, I’ll do some kind of rudimentary query and it pretty quickly I get some very confident, incorrect answer. And it’s like, okay. so yeah, I think that’s a fair worry.

Taneth (39:52.338)

And I mean, of course, you know, the internet isn’t built for LLMs, it’s built for Google, you know? And so, of course, it’s going to get things wrong. We’re kind of not helping it right now. I do think it will improve as the output is only as good as the input. I think the input right now, as in the entire World Wide Web, is like not structured for this. So of course, that’s going to happen. I do think it will improve. And I do think as we all learn,

Pete Pachal (40:05.902)

Mm.

Taneth (40:22.024)

good stuff in, good stuff out, we’ll get higher quality answers. But that will be contingent on a lot of education and making sure that everyone kind of has access to the right technology and information, I think. Okay, on brighter note, I’m really hopeful that this technology will allow

Pete Pachal (40:41.144)

Nice. Yeah. What’s the thing you’re hopeful about?

Taneth (40:50.716)

us to deliver information, good quality facts, news and information to more people. Because I hope that more people will want to interact with the kind of news and information that we’re delivering them because we’re doing it in a more effective manner. I spoke earlier about younger generations turning away from news. And I fear that I think it’s

really, I think it’s an emergency that we are creating things that are relevant to generations in a way that they want it. And I really think that AI is going to help us do that in a way that we’ve never been able to do before.

Pete Pachal (41:31.82)

Nice. We’ll leave it there. Tenneth Evans, thanks for coming by and sharing your thoughts.

Taneth (41:35.912)

Thank you so much.

The post The future of journalism is personal: How The Journal is building AI for readers, not robots appeared first on The Media Copilot.

The Fable 5 pullback turns AI availability into a planning problem

Pete Pachal — Tue, 23 Jun 2026 12:00:00 +0000

The AI industry pumps out so much hype that you’d be forgiven for simply shrugging at the recent release and sudden withdrawal of Anthropic’s Fable 5 model. Set against the Elon Musk vs. Sam Altman trial and Anthropic locking antlers with the Department of War, the Fable episode could read like just another week in AI.

This one is worth paying attention to. This is really the first time the government has stepped in to regulate a specific model release on the grounds that its capabilities could pose a national security risk in the wrong hands. Whatever happens next, the line has been drawn: a frontier model in general release can be taken off the board because Washington decides it’s too dangerous to leave widely available.

For anyone building AI into their daily work, that shifts the calculation in a real way. The intelligence available to you isn’t only a function of price anymore. It’s also a function of policy, geography, the terms you’re willing to accept on your data, and whether the vendor or the government leaves the model running at all.

The story behind the freeze

For readers who don’t track model releases closely, here’s the short version. Fable 5 is the first generally available model in what the company is calling its “Mythos-class” models, a tier above Opus that Anthropic says has crossed a meaningful risk threshold in cybersecurity and biology. Fable 5 is the consumer-safe version, built on the same underlying Mythos 5 model but wrapped in extra guardrails designed to block or downgrade certain cyber, biology, chemistry, and model-development queries. It also jumps Anthropic’s core model number, signaling a generational step forward from Opus 4.8, Sonnet 4.6, and Haiku 4.5.

Then, on June 12, three days after launch, the government ordered Anthropic to block Fable 5 and Mythos 5 from every foreign national, including foreign-national employees working inside the United States. Anthropic said it could not reliably enforce that distinction and disabled both models globally. The trigger, by most accounts, was a suspected jailbreak that punched through Fable’s cybersecurity guardrails. Anthropic disputed the severity of the finding, saying the demonstration uncovered only minor, previously known vulnerabilities that other public models could also identify.

That fight is still going on. Cybersecurity leaders have urged the government to reverse the order, arguing that defenders need access to the same capabilities and that comparable tools are already available from American and Chinese competitors. Anthropic is working to get Fable back online, and rival labs will almost certainly ship something comparable in short order (some are already claiming to have done so).

The specific dispute may resolve in days or weeks. The precedent will outlast it. A model can be released, integrated into workflows, and then disappear because a government draws a line around who may use it. For anyone building around a single model or vendor (and “building” might simply be leveraging it in crucial, strategic use cases), availability is now part of the risk calculation.

What people saw before the lights went out

Early users got just enough time with Fable 5 to confirm Anthropic’s claims about it. Despite controversies over how Anthropic chose to limit how Fable 5 deals with queries the company deems risky (more on that in a minute), users are seeing the power of the model. Fable 5 is designed for agentic work, meaning it can work autonomously on tasks for a long time, sometimes hours or days, without losing context. The advice that came out of those early sessions was consistent: stop using frontier AI like a fancy autocomplete. The best way to use it, many say, is not to ask it to perform straightforward in-and-out tasks like writing an essay or telling you the best parts of a lengthy report, but to give it broader goals about what you’re trying to achieve, let it build the plan, then execute, however long it takes.

That window was short, but it counted. It showed this level of intelligence is no longer a slide in a research deck. The model was pulled back, but the capability threshold remains crossed.

A big part of what makes Fable work is that it grades its own homework. If you’re a regular user of Anthropic’s models, you’ll notice there’s no “Thinking” mode for Fable 5. That’s because adaptive thinking is always on: The model decides when and how much to reason on every request, and at higher effort levels it can reflect on and validate its own work. Tasks turn into loops. As it works to achieve the goal, it can try things, evaluate the results, change course as needed, and try again. And it can do so autonomously.

For media and marketing teams, the practical shift is in scope. Instead of, say, assigning it to design a specific email campaign, or help format your newsletter, you can zoom out and tell Fable 5 to conceive and build an entire marketing strategy around your newsletter. That might involve reformatting your templates, building new landing pages, adjusting the publishing schedule, building a social campaign, and more. Theoretically, with the right access, it could then build all of that for you. Your job is to grade the output. Over time, less of that grading happens mid-process and more of it happens at the end.

That’s the promise anyway. The danger is that organizations may begin designing around that promise before access, cost, and governance are stable enough to support it.

Fable 5 is the first model that puts real agency on the table. Right now, working with agents, while powerful, involves a lot of management: ensuring the plan the agent builds is correct, clearing up barriers that it encounters as it performs the task, and then guiding it to the best output, usually through multiple iterations on the task itself. In theory, a model strong enough to evaluate its own intermediate work shouldn’t need that hand-holding.

That gap between theory and practice is the real story of the freeze. For a few days, users could test a different relationship with AI; then the capability vanished. We crossed the threshold in the lab and lost it in the market on the same week.

The three walls between you and frontier intelligence

Fable 5 and the models that will follow it stand to change how we work with AI, and arguably how we work, full stop. However, using Fable 5 to its full potential was never just a matter of selecting it in your model picker or calling the API and letting it cook. The pullback put a sharper point on a problem that was already there: the most capable models are also the hardest to actually deploy. I see three walls in the way, with a fourth that just got built.

Access and context. For an organization to use Fable 5 to its full potential, it would require a large amount of access to the right context (the org’s information and data). Here, Fable’s strength tripped over itself. Because Anthropic fears the model could be misused, it requires prompts and outputs from Mythos-class models to be retained for at least 30 days for safety monitoring, including in enterprise environments that would otherwise use zero data retention. Anthropic says the data will not be used to train models and that, on some third-party platforms, it remains inside the customer’s cloud environment. But companies cannot use Fable 5 under a true zero-retention arrangement.

That retention requirement, plus the restricted categories where Fable 5 quietly throttles down to Opus 4.8, has set off real friction with enterprise buyers. Many companies will be reluctant to cede control over how their own data is retained and reviewed. Microsoft reportedly limited employee access while its legal teams assessed the implications for confidential and customer data.

And on top of all that sits the new wall. Even if a company accepts the privacy terms, secures the integrations, and builds the right internal controls, the model can still disappear because of a government order or vendor decision. Serious agentic systems will need fallback models, portability across vendors, and a plan for what happens when the most capable model is suddenly unavailable.
Compute. Fable 5 is not cheap. Anthropic priced it at $10 per million input tokens and $50 per million output tokens, twice the price of Opus 4.8. I’ve written before about how the agent era is squeezing compute budgets at every layer, and with AI hardening into a political wedge issue, expect compute pressure to stay tight for months and probably years.

The premium price doesn’t automatically kill the math. Some early users argued that it could solve hard tasks in fewer turns than weaker models, potentially lowering the total cost of completing the work. Still, that argument only holds if the work was worth doing with a frontier model in the first place.

If Fable 5 and its peers are going to act as the brains at the top of a company’s AI stack, the deployment question is going to need actual rigor. Organizations will need to be very selective of how to deploy it: which tasks to assign to it, who should have access, and what guidelines, rules, and restrictions there need to be on usage.

And there’s an awkward irony in talking about allocation right now. Intelligence can be technically achievable and commercially valuable while still being unavailable.
Task imagination. I became aware of the term “task imagination” through the AI Daily Brief podcast, which references a video by the AI strategist Nate B. Jones. In his take on the Fable 5 release, Jones makes the simple observation that not many knowledge workers think about their work in terms of tasks that might take days to do. It requires a certain level of strategic thinking that may not actually apply to many roles. Put bluntly: a model can run for two days, but most workers have never been asked to define a goal worth two days of machine effort.

For media practitioners, that’s the part worth sitting with. An editor might call on the model to develop more granular editorial guidelines and style guides based on different article types (news, features, evergreen explainers, etc.). Reporters might build investigative agents that don’t just surface data in document troves, but develop research plans based on leads and then execute on them by mining remote databases, filing FOIA requests, and other complex touchpoints that typically require human involvement.

The catch is that most jobs aren’t scoped that way. Many jobs have narrow definitions of what the work is, and there’s little motivation to go beyond that. A model that can do days of work isn’t very useful if the work it’s given is still measured in minutes. That puts pressure on workers to imagine more ambitious tasks or risk being left behind.

The paradox of a pause

The Fable 5 pause comes wrapped in a paradox. A pause gives organizations time to build the governance, data practices, and strategic habits needed to use this level of intelligence responsibly. The trouble is, task imagination only develops with hands on the model. Without access, people cannot discover which long-running assignments are worth the money, where agents fail, or how their own roles could expand around them. The pause buys time while taking away the main way to use that time well.

Step back, and a clearer picture forms. A future where we’re working alongside agents will encounter serious barriers beyond just capability (and political freak-outs over that capability). We restrict access to context so neither the tool nor its creators knows too much. We limit how much we spend on models because we’re unsure of the return we’ll get. And many of us throttle our ambition with AI since our jobs simply don’t have a rich enough canvas for a model like Fable 5 to fill in.

A fourth restriction now sits on top of those three: the model itself may simply not be available.

For media leaders trying to make the ROI case, that’s a problem. The strongest demonstrations depend on giving capable models real work, real context, and enough time to execute. When the most capable models are pricey, hemmed in, or suddenly absent, teams drift to safer pilots that are easier to approve and unlikely to move the underlying economics.

None of these walls fall just because someone ships a smarter model. While advancements in security, infrastructure, and work redefinition will help us get past them, those are inherently slower than the rapid advancement of AI.

We pushed past one threshold and walked straight into several walls. I suspect the story of Fable 5 will be looked back on not primarily as a step up in power, but as the moment where the implications of that power pushed the limits of the systems meant to use it. Agentic AI is clearly where this is going. The systems around it need a beat to catch up.

The pause is useful, but it isn’t free. Experimentation is how organizations learn what this intelligence is actually for. For now, AI leaders are about to discover that running frontier AI at full strength is harder than proving the strength exists. The pure experimentation phase is over. The reality check phase has started, and access, cost, control, and utility now matter every bit as much as raw intelligence does.

A version of this column appears in Fast Company.

The post The Fable 5 pullback turns AI availability into a planning problem appeared first on The Media Copilot.

AI’s reality check: Why Sharon Goldman is looking beyond the hype

The Copilot — Thu, 18 Jun 2026 04:00:00 +0000

This episode is sponsored by: Adobe Acrobat

This week on The Media Copilot, Pete Pachal sits down with veteran AI journalist and former Fortune reporter Sharon Goldman to discuss the growing disconnect between AI’s promise and public perception. As Sharon launches her new Substack, Ground Level AI, she shares why she’s shifting her focus away from model releases and Silicon Valley headlines to examine how AI is impacting communities, businesses, governments, and everyday people.

“To say you’re covering AI today is like boiling the ocean.” — Sharon Goldman

From AI data centers and cybersecurity risks to job displacement fears, media disruption, and public trust, Sharon offers a grounded perspective on where the AI conversation is headed next.

Listen or watch:

YouTube

Spotify

Apple Podcasts

What we cover

• Why AI companies are struggling to win public trust
• The growing backlash against AI and what’s driving it
• How data centers, infrastructure, and policy are becoming major AI stories
• Whether AI’s impact on jobs is being overstated
• The future of journalism in an AI-powered information ecosystem
• Why independent voices matter more than ever in technology reporting
• How Sharon uses AI as a reporting partner, editor, and research assistant
• The biggest AI stories to watch heading into 2027

If AI is reshaping society, who gets to tell that story? Sharon Goldman believes the most important AI stories are not found in the latest model release or product announcement. They are happening in communities, workplaces, governments, and everyday life, where technology is creating real-world impact.

Link: Do that with Acrobat: AI-Powered PDF workspaces | Adobe Acrobat

Why this matters

AI is no longer just a technology story. It is influencing how businesses operate, how information is distributed, how governments make decisions, and how communities adapt to rapid change. Understanding AI’s real-world impact is becoming just as important as understanding the technology itself. As adoption accelerates and public skepticism grows, the conversation is shifting from what AI can do to how it affects people, jobs, infrastructure, and society as a whole.

About the 👤 Guest

Sharon Goldman on LinkedIn

Ground Level AI

Enjoyed this episode?

Produced by Pete Pachal and Executive Producer Michele Musso
Edited by the Musso Media Team

Music: “Favorite” by Alexander Nakarada, licensed under CC BY 4.0

TMC- TRANSCRIPT SHARON GOLDMAN

Pete Pachal (00:19.51)

Hi, welcome to the Media Copilot. It’s a podcast about how AI is changing media, news, and communication. I’m your host, Pete Paschel. I covered tech for a long time as a journalist, and now I have deep conversations with media people, with builders, and with creators who are all answering the question how will we get information in the future? And how will that change the jobs and the industries whose business is information, especially media? My guest today is Sharon Goldman.

Sharon has been one of the most thoughtful reporters covering AI, most recently at Fortune, where her work has gone well beyond product launches and model rankings. She’s covered AI as a business story, an infrastructure story, a labor story, a security story, a policy story, and increasingly as a story about power. Who controls the systems, who pays for them, and who has to live with the consequences. Now Sharon’s moving into a new chapter. She’s leaving Fortune to launch ground level AI.

A publication focused on AI as it meets the real world. Infrastructure, geopolitics, societal issues like jobs and the environment, all the messy parts that don’t fit neatly into a press release. So today we’re going to talk about what she’s learned from covering AI up close, where what she thinks the industry still gets wrong, how the public’s responding to AI, and what all this means for media, journalism, and our information ecosystem before we get into that though, please just take a second to rate or review the show if you can. If you’re listening on Apple or Spotify, please leave a five-star review. Or if you can, maybe a nice comment. And if you’re watching on YouTube, please like the video and subscribe to the channel if you don’t mind. Those things really do help people find the show. All right, housekeeping over. Sharon, welcome to the Media Copilot.

Sharon Goldman (02:10.748)

thank you so much, Pete. Thanks for having me.

Pete Pachal (02:14.237)

Awesome. So before we get into all the things that I was promising there, I’d love to hear just a little bit more about you, your background. I know you’ve been covering AT A AI for quite a while, not just at Fortune, but other publications. Tell us a little bit about your background, how you got into journalism and how you got into AI reporting.

Sharon Goldman (02:30.584)

Well, as far as getting into AI reporting, that started in April 2022. So a little bit over four years ago, I was hired at a publication called Venture Beat. And the week that I started, the person covering AI left to go to TechCrunch. And I raised my hand and I said, Well, I’ll cover AI full-time as my daily beat. And that was six months before ChatGPT came out, and I’ve been on a roller coaster ride ever since.

Pete Pachal (02:55.533)

Well timed.

Sharon Goldman (02:58.776)

As far as my journalism career, I’ve been in the business for over 25 years. I started out in B2B journalism in subjects completely unrelated to technology. But about 15 years ago, I did start covering technology from the standpoint of marketing tech and sales tech, when Salesforce first came into the picture, when SEO was new, when social media was new. So I feel like I’ve

Pete Pachal (03:26.389)

Well those are the days.

Sharon Goldman (03:27.788)

Those were the days and I feel like I’ve constantly been on this trajectory of of being at sort of on the front lines of when some of these technologies got started and that always floats my boat. So it’s been a really fun and exciting journey that I’m thrilled to continue.

Pete Pachal (03:47.266)

Yeah, it’s pretty wild to know s note how much all of that stuff has changed as you were sort of describing all those areas. And AI, of course, is like always changing. Like I feel like, you know, everything changes, but like, holy cow, is it so different from when you first started it, even in 2022? So, you know, tell us about like what you’re doing now, because you’re leaving fortune to start this new publication, ground level AI. What made you decide like this was the moment here in 2026 to do that?

Sharon Goldman (04:13.196)

I feel like there were a few things converging for me. For one thing, it’s kind of in my DNA to be a bit of a builder. I’ve always had an entrepreneurial bent. Before I started covering AI as a full-time daily beat for venture beat and then fortune, I actually was a freelancer for over 10 years. and I was very successful doing it. I really loved it. I loved that kind of autonomy and sort of building a business on my own.

So I feel like it’s a little bit going back to my natural habitat in that regard. I also feel like the timing is really right. This is such a momentous societal shift that deserves all kinds of coverage. You know, I feel like a lot of coverage right now, you know, really hones in on the biggest companies, the biggest models, you know, the the the gossip, the drama.

But you know, I’m really interested in digging into it’s such a broad beat at this point. Like to say you’re covering AI is like boiling the ocean. So I feel that there’s a gap, you know, in being able to narrow that down to some of the things I’m most interested in. And, you know, after, you know, over two decades in the industry, I feel like I have a voice, I have a point of view that I’d really like to.

get across. And I finally I feel that there needs to be more independent female tech voices out there in the independent journalism ecosystem. And you know, I’m definitely ready to jump in on that.

Pete Pachal (05:48.77)

Nice. So you described, I think, ground level AI as like it’s AI meets the real world, right? And what what does that mean in practice? It seems like that’s a good reflection, I think, of like a lot of your coverage of fortune, which talks about like, you know, these ground level consequences of data centers, the skepticism around it, et cetera. are are you gonna continue in that vein? does it change at all? Do you add to it? How is how is ground level AI gonna cover AI in the real world.

Sharon Goldman (06:19.862)

Yeah, so AI in the real world can can seem a bit vague. To me, it’s really about this societal shift that we’re seeing right now. And that, you know, is physically when it comes to data centers, and we’re seeing, you know, a tremendous build-out, but also a tremendous backlash. And when those two things converge, you know, there’s a lot going on. I also see it as far as you know, just policy, how governments and organizations are going to handle this shift going forward when it comes to issues like labor, when it comes to issues like safety. You know, that’s everywhere right now. With a big election coming up in November, I’m really eager to dig into that. I’m also keenly aware of security issues that are every single enterprise is needing to invest in this right now with anthropics mythos.

You know, kind of changing the game and giving enterprise companies a wake up call about what they need to do in their organizations to make their systems secure. I think that’s the real world too. That’s ground level change and that’s what I really want to dig in into.

Pete Pachal (07:36.344)

Yeah, and also feel like you mentioned the election there in terms of AI meeting the real world. You know, I feel like that’s actually happening in a sort of mainstream consciousness way, right, in this moment too. And I’d love to to get your thoughts on, you know, the public’s relationship with AI, which, you know, certainly in recent months has seemingly taken a turn for the negative. there’s been a lot of shall we say skepticism about AI that’s gone mainstream. You know, there’s the the infamous graduation speech I guess trend that that where people would mention AI and get booed. I’d love to get your sort of initial thoughts on that and let’s let’s sort of double click on a few.

Sharon Goldman (08:18.86)

I feel like I’ve been beating this drum for quite a while. A few years ago I was starting to write essays and and analysis pieces where I the communications piece has got to get better, I think, from the AI company standpoint, if they want the public to get on board. You know, you can’t, in my opinion, say in the media that you know, half of all white collar jobs are gonna go away by this time and then expect everyone to be excited about the photos and images and and writing that they’re doing using their tools. So I feel like there’s a tremendous disconnect. you know, and then where is the public getting their information from? You know, they’re certainly getting news from social media and there’s a tremendous backlash on social media to AI, whether it’s from the data center standpoint, whether it’s job loss. Also just I feel like there’s a bit of a disconnect, partly because it’s how the technology is. It’s these use cases now, whether it’s writing or coding, these are not the world-changing use cases that people would like to hear to think that they have to suffer for it.

If AI really was, for example, curing cancer, well, that would be a different story. But if it’s simply making your enterprise workflows better or creating an app, people commonsensically are not going to see that as potentially worth it. But of course, the other side of that is that AI is an exciting technology with a tremendous amount of potential for humanity.

But I don’t think that’s being communicated very well. And from a politics standpoint, that’s going to be big in November.

Pete Pachal (10:16.257)

Yeah, it’s really interesting like the communications aspect of it that you touched on there because it here’s a little theory I have. I’d love to hear your thoughts on it, in that a lot of I obviously cover media and AI all the time, and there’s a lot made of how the media is now reacting to AI with the knowledge of what happened in search and social and that it all eat up their business model. So there’s a very sort of defensive view it has generally. But there’s a flip side to that.

Right. There’s and this is I feel like this is what may have sort of guided the comp, like whether it’s a strategy or whatever, like the I because here you here’s again the bear with me here. But like the jobs stuff, let’s just take that. They were they were crowing about how AI was going to kill all the jobs, like people like Dario Amade and everybody, before there was data, you know, and and so far to my knowledge, again, you correct me if I’m wrong, because you’re on top of this more than me.

I don’t think there is data yet that they’re that AI is like massive job loss. You know, there’s the anecdotal stuff, of course, but like the it’s just not there. So it’s a weird thing that they decided to come out ahead of it as sort of a comms tactic. And I feel like their lesson from that era is like, let’s talk about the negatives before they become an issue so that we’re not, you know, pilloried for it later or something. Was that

Do you think is that a theory? Like what do you think? Like is I to me it confuses me why they’re the worst pitchmen for their own tech because they they’re like kill saying it’s gonna kill everybody and before it there’s even any data to support that.

Sharon Goldman (11:48.706)

That is my theory. I I remember a few years ago, Sam Altman saying something like, you know, we’re we’re putting this technology out so people can get used to it, so they can kind of understand and in and embrace and experience what’s going to happen before before it goes further, before it gets even more advanced. And in a way, I felt like the job loss idea is sort of jumping ahead and saying, like, you need to beware, you need to reskill, you need to think about.

This, but again, before really any data has come forth. Also, they’re really thinking about within their own fields. I mean, when it comes to tech, and you know, if you’re a software developer, yes, there are job losses, there are layoffs in tech companies. This is happening to some extent, but that doesn’t broaden out necessarily to every field every position. And it also doesn’t take into consideration the potential for job creation, which, for example, Aaron Levy at Box has really been talking about beating that drum on social media lately. that new jobs will be created, that agents are still going to need humans to to manage them, to, you know, and and jobs that haven’t even been thought of yet. That’s what I wonder about.

You know, my husband is in cybersecurity, for example. Well, cybersecurity wasn’t a thing until, you know, twenty years ago, twenty-five years ago. There are there are many jobs like that. there were no social media influencers 15 years ago. So who knows what’s going to come. But I do think that you know, I guess that frontier labs like anthropic and open AI kind of with their they’re they’re also coming with their own baggage of what they talk about as far as AGI, what they believe is going to happen in the future. But again, the timing of that, is that going to happen in two years? Is that going to happen in 20 years? you know, I I just don’t think that that’s the way. Or maybe they just feel like they didn’t have enough to say yet about the positives. I don’t know.

Pete Pachal (14:03.703)

Right, right. And I feel like like you mentioned like of going back to the public sentiment, like are everyday people experiencing the positives. Yes, of course, if the AI could hear cancer tomorrow, we’d all celebrate. That’d be a huge thing. It hasn’t yet, notably. But the the the idea that like it’s making your life easier, better, etcetera, it’s to to me to my like the thing is here’s the thing. You and I I assume you’re kind of a an adopter and a trier, you know, you use this technology too. And we should get I’d love to hear about like how you’re using it. But I’m seeing good gains because I’m a power user, right? And I’m diving into these things all the time. And there’s co-working apps. And it’s like, this so cool. I think that’s very rare. I think we’re like in the story like the 1% of people and it’s a ways out. So this sort of it is anti AI stance. I feel like it’s hard to pin down on it’s not just jobs. It’s not just environment. I think it’s just this broad like

Sharon Goldman (14:32.994)

Yes.

Sharon Goldman (14:42.87)

Yeah.

Pete Pachal (15:00.917)

what’s so great about it sense that sort of people have about it. And then, yeah.

Sharon Goldman (15:05.26)

Yeah, these are not, these are not the power users. Excuse me. These are not the people necessarily even using it at work. you know, what you you talked a little bit about search and social media. I mean, these were things that really were kind of life-changing at the time, like the idea that you could, you know, follow a friend or message a friend or search for people it, you know, that you knew decades ago that you weren’t able to keep in touch with.

Pete Pachal (15:21.014)

Right.

Sharon Goldman (15:33.218)

That was a real life-changing thing, even if you feel that today there are so many downsides. That was a real game changer. Here with AI, I feel like it’s much more, it’s sort of om it’s always sort of what’s coming, what’s coming down the pike. It’s not necessarily what’s here and what you’re playing with it now. It’s also something that does take some doing as far as understanding how it could work for you.

Power users like you and I, we we all spent a lot of time trying to figure out, well, what is this really good for and what is it not you know, fun but not really useful.

Pete Pachal (16:12.342)

Yeah. What do you think AI companies should do in this front in terms of the massive wave of public s skepticism now? And you know, we can touch on the environmental stuff too, if you like, ’cause I feel like that’s a huge chapter of this. And I know you’ve covered that quite a bit. But what are what are the things they can do about this to maybe turn the tide on the sentiment, but at least between now and maybe the election? Well, not not that that necessarily favors one party or the other, but maybe it does. You tell me.

Sharon Goldman (16:34.712)

Right.

Know. I mean, these these companies now have massive comms teams, you know, that that that just are overwhelming. Hundred I remember when OpenAI just had one comms guy and now there are hundreds and hundreds. I do think, you know, I mean, and maybe they feel like they have been doing some radical transparency, but you know, transparency is key. You know, when I report on AI data centers, for example, it’s one of the things that communities

Pete Pachal (16:43.842)

Yeah.

Pete Pachal (16:50.702)

Mm.

Sharon Goldman (17:06.172)

Are so frustrated by kind of the kind of the secret nature of it that things just appear. There are NDAs and things you can’t talk about, and then suddenly, you know, something is appearing before their city council. you know, I think also just you know, maybe empathizing more and not being so Silicon Valley-centric and assuming that everyone is so excited about the the the deep tech or the…

Pete Pachal (17:21.71)

Mm.

Sharon Goldman (17:35.692)

… use cases that require you know several hundred dollars a month in tokens. I mean that this isn’t you know really speaking to the average person and and also making it clear that you know this is this is our opinion, this is the way we see the world, this is the way we see things playing out, but there are other opinions out there. There are other people saying different things.

Pete Pachal (18:03.982)

Let’s switch gears for a bit and sort of talk about the media for a bit, because that of course is what I what I zero in on a lot. well let’s just talk broadly first, actually, about the information ecosystem and stuff. Like what do you think is the biggest thing in your mind about how AI is changing that? How AI is changing what we’re doing? I I assume, you know, vis-a-vis your new venture is that you’re gonna have some kind of website and and how you’re feeling about that in terms of what AI is gonna do to it and et cetera. Maybe you even have some thoughts on whether to block or not block or what have you. But you first tell me like broadly like how are things changing and then maybe let’s make it a little more personal.

Sharon Goldman (18:43.458)

Well, ground level AI is gonna be on Substacks, so that’s one thing. And I’ve already had a sub yes, that’s still the internet. yeah, I mean, I think that, you know, as we all know, the media ecosystem has changed already almost beyond recognition. I think that you can’t you you just can’t rely on people finding you through search. You can’t rely on on

Pete Pachal (18:46.454)

Okay. Well that’s still the internet.

Sharon Goldman (19:12.576)

Google to save you. I think that the that that we’re all you so many media companies are using AI that we all have to think about our voice. We all have to think about what makes us unique and what we can offer that goes beyond the chatbot.

Pete Pachal (19:33.921)

And like the whole idea of okay, people get your information through a summary now, or at least that’s sort of a discovery window of some sort. and I don’t know like how much, you know, you can tell me about like how fortune sees that and if they’re doing things like GEO or what have you. but how is sort of covering this space sort of as you’ve thought about just even even yeah absent your new adventure, whatever, like I’m reaching an audience now through these this sort of you know disintermediation layer, the summarization layer. And

Sharon Goldman (20:05.688)

I actually remember I remember speaking to someone at a conference, a creator, and the creator was telling me, I think this might have been like three years ago, that this person wanted to be discovered by AI. He wanted Google and OpenAI and anthropic chatbots to find him and scrape him and spit him out because it was gonna help other people find him through GEO, which wasn’t even a term at the time.

Pete Pachal (20:19.758)

Mm.

Sharon Goldman (20:35.244)

And at first I was like, wow, that’s amazing that a creator would say, I want to be scraped. But today I think that is real and true. And I would have to agree with that. I don’t think there’s any point in saying, like, you can’t scrape my work, you can’t output, you know, something related to me and you know, link to my work through through you know, Claude or Chat GPT.

Pete Pachal (20:35.288)

Right.

Sharon Goldman (21:01.9)

I think that’s happening. And I think that ever every single brand out there is working towards it. I’ve noticed in my inbox several companies that this is what they do. They rep, they they work for brands and they help them get discovered. And it’s a new, it’s a new way to discover. And in my own use of AI, even outside of my professional life, I use it for shopping, I use it for research. I use it for all sorts of things. So I want to discover things through.

Claude and ChatGPT and Gemini in the same way that I would be looking for things through Google search.

Pete Pachal (21:37.507)

Totally. And you d it helps it’s very helpful when it can just bring you that information. but of course it needs to get that information from somewhere, it needs to get the context, and it ingests things like, you know, obviously journalism and other content and gives you the answer. A. G. Sulzberger just recently made a speech at a conference and he was saying that, you know, this is essentially theft. This is like if you’re a may a may building a tower, a construction company, but you get all the materials from the the village next to you or something and you just don’t pay for them. you know, again, I I don’t know how you see it, but if the copyright question is a is a is the thing that’s out there. from your reporting, I guess what I what I’m really curious about, and I don’t know if you’ve covered this issue specifically, but from what you know about the AI companies, like how seriously do they take that point of view and concerns about copyright, you know, vis a vis the

I know they have lawsuits and stuff, but like how do how do you what is your sense of like how they even think about this?

Sharon Goldman (22:40.118)

The AI companies believe that what they’ve done or or they, you know, they their their PR and strategies to communicate that their belief is that it’s fair use. You know, whether that was in their early, you know, whether that’s what they thought initially back in the day, or whether it was more like we’re we’re using this for research and it won’t matter in the future. But at this point, you know, I’ve spoken to OpenAI’s lawyers, for example, and their argument is that it is fair use. You know, this is this was you know, vast swaths of of public material. And even if there is some copyrighted material in there, you know, this is about the output, and the output isn’t the same. And you know, that’s that’s their argument. I actually thought that more of these lawsuits.

From the standpoint of the defendants, the author, the the, I’m sorry, the plaintiffs, the people bringing the lawsuits, authors, et cetera. I had a lot of predictions from lawyers saying that this would reach the Supreme Court, that, you know, that that there was a really good argument there. But I do think the AI companies have, you know, in incredible legal representation and the argument around that what they’re doing was fair use.

Has gotten a lot of traction, I think.

Pete Pachal (24:11.278)

D interesting. So do you think it will end up getting the Supreme Court? I guess ultimately, like, how do you think this question will be settled, at least in the United States? I mean, we’ve seen like o in Europe, for example, there’s well, you know, there’s there’s a different stance, shall we say, and there’s probably a little bit more movement to protect content creators. The UK body just re this is a little different issue, but the UK body recently sort of is gonna be forcing Google to separate their AI crawler from their search crawler so that copyright holders can opt out of AI search, which they kind of can’t do right now with respect to Google. kind of an in the weeds thing there, but like the broader question, how do you think this is gonna be resolved?

Sharon Goldman (24:50.744)

I feel a little

Sharon Goldman (24:55.532)

I feel a little bit cynical there. I I kind of feel like already we’re seeing the copyright issue not as front and center as it was a couple of years ago. I feel like a lot of the arguments have been thrown out by judges and there are some left, but I I I think the the the fact that copyright laws were not written for this age is the big the biggest problem. And so potentially there isn’t a way for a court to say that that OpenAI and Anthropic are completely unable to do this. And I do think that these are two massive companies now with incredible investments in their legal teams, and they have also a strong argument. So I guess cynically I kind of feel like they’re winning on the copyright side of things.

Pete Pachal (25:58.223)

Staying with like the information side of things, I know you’ve covered like the content credentials and C2PA and these standards to authenticate or potentially authenticate like what’s an AI image, what isn’t? I’ve looked at this a little bit too. it’s it’s a little depressing how l how little traction it seems to have gotten so far, at least as far as I can tell. What is your what are your thoughts on this? Does this have a chance of helping us?

kind of detect what’s r real, what’s not out there, is which is obviously a big issue as as more AI content gets distributed.

Sharon Goldman (26:38.04)

I also am a little cynical there. I think that it might help if you like in an individual scenario where you needed to find out where something specifically was real or not. But as far as like the vast swath of slop out there and your average person being able to discern that something is real or not, I think the answer is no, even if there is a symbol or whatever, something can get shared very widely before that’s even noticed or you know whether people would even believe those markings is is another issue. I think when when I’ve spoken to some experts in this field one thing I thought was interesting is the idea of doing the opposite rather than proving something is fake proving something is is real you know I f I feel like

It’s hard to say whether any of it will work though, because I feel like there’s there’s such a vast swath of slop now.

Pete Pachal (27:44.502)

Yeah. Th there definitely is. how do you define slot, by the way? Is it just you anything AI generated or is it anything AI generated that’s either worthless or misinformation or what’s what’s your general definition of slot?

Do we need one? I feel like slop has become fake news, you know what I mean? Like it’s like it kinda means what we kinda know what you mean, but kinda it’s there’s not really a good definition for it.

Sharon Goldman (28:09.634)

There isn’t. And I’m not sure all slop is slop. Like, for example, I’ve been thinking a lot about the political videos that have been going out. They’re AI generated. you know, with Trump, for example, or I’m thinking of like Spencer Pratt, the the LA mayoral candidate putting out videos. Like, are those slop? Like they have a purpose. They’re

Pete Pachal (28:21.504)

Right.

Sharon Goldman (28:35.788)

They’re being used for political gain, they’re sending a message. Is that slap in in

Pete Pachal (28:36.013)

Right.

Pete Pachal (28:41.484)

I feel like slop to your point connotes l low value. And while those are clearly AI, they’re valuable to somebody, certainly valuable to the campaigns or they wouldn’t be doing them. and so I don’t I don’t think it does. I don’t think it it meets the bar on slop, even though it’s so obviously

Sharon Goldman (28:54.06)

Right.

Sharon Goldman (29:01.1)

Yeah, I feel like slop is sort of, you know, vast quantities of low value, c clearly rather useless content, whether images or text.

Pete Pachal (29:12.686)

Mm.

Pete Pachal (29:17.442)

So I’d love to talk about, you know, obviously you’ve been reporting on this for a number of years. You have thoughts on just the how to report on this, which is to say, like, what are the biggest lessons you’ve learned? What makes AI reporting challenging? How have you been able to cover it well despite those challenges? Especially as I’m sure you’ve probably seen the number of AI reporters multiply over the past the past few years.

Sharon Goldman (29:43.288)

Well, absolutely. When I first started covering AI in twenty twenty-two, there there was a community of AI reporters, but it’s multiplied exponentially. And in a way, you know, at Fortune, for example, in a way, everyone is an AI reporter because almost everything we’re reporting on, so much of it has from a business standpoint, has an AI angle. I feel like the biggest lesson I’ve learned covering AI is.

How it has the news the news cycle is so fast, so speedy. What was news yesterday might not even be news anymore the same day. And the amount of news that just comes flowing like a tsunami on a daily basis is impossible for one person to cover. So to say you’re an AI reporter today is, you know.

Not the same as the AI reporter from a few years ago who was reporting on a few companies, frontier companies that were mostly research labs that were putting out groundbreaking research, or they were focused on enterprise use cases that were very much under the hood, that were, you know, unsexy. Today that’s completely, you know, the the amount of startups, the amount of funding, the amount of investment, the the idea that it’s propping up the entire US economy and global economy, the idea that it’s become such an issue of geopolitics between the US and China that it could actually be dangerous is you know, that’s a whole other thing. That requires dozens of reporters to cover every angle and every company.

Pete Pachal (31:25.166)

Mm.

Sharon Goldman (31:34.136)

So I feel like the lesson I’ve learned is to take a step back and be like, well, you know, not what’s the news today, you know, whether it be another embargo or another piece of gossip from OpenAI or Anthropic, but what’s most important to cover? What’s the biggest message to respond to today or to communicate? Or what am I most interested in in covering today or this week?

That you know, I have to step back, you know, for mental health purposes as well as what I see happening in the in the landscape, and say, what what’s really important here for people to know? and a lot of times that isn’t the biggest model or the latest release or the the biggest valuation. It’s really about what’s happening underneath, you know, it’s the whether it’s the

Pete Pachal (32:08.994)

Sure.

Pete Pachal (32:17.326)

Yeah.

Pete Pachal (32:30.712)

No.

Sharon Goldman (32:32.652)

the nuts and bolts of the infrastructure or the safety or what’s happening to communities or what policies are being implemented, what people are talking about behind closed doors. Those are all really important.

Pete Pachal (32:47.414)

Yeah. Do you feel like that’s kind of undercovered these days? Like do you use is there too much in the hype cycle, the products, what these things can do, or ’cause I feel like when I was covering tech, there was always like and I was covering it more from the consumer standpoint, but it was like the stuff would come out, it was pretty easy to get get pulled into like, you know, the reality distortion fields, the hype cycles and get speculative. Yeah.

Sharon Goldman (33:08.522)

Exactly. I I feel I just feel like the nature of it is it’s happening so fast. I mean, I think you know, last month maybe Anthropic had had dozens of releases, you know, you can’t possibly be covering each one you know, as as a small tech team of reporters, for example. so you you you kind of have to get outside that hype cycle. And I do think it’s undercover. That’s part of

Pete Pachal (33:21.197)

Hm.

Sharon Goldman (33:37.677)

what is driving my desire to launch ground level AI is to is to turn my focus to that and leave some of the you know the coverage of the latest drama or personnel move or even you know the latest model change from five point five to five point six to someone else.

Pete Pachal (34:01.336)

Right. Yeah. It’s always like some increment. It’s and it seems like it’s a big deal when the whole number changes, but then it kinda isn’t. It’s weird. I feel like AI, the industry just decided they have they’re not even gonna try to fix their naming slash numbering thing.

Sharon Goldman (34:15.008)

No exactly. The for forget fixing the naming conventions. We’ll just carry on.

Pete Pachal (34:19.094)

Yeah. Okay. Yeah. I teach this stuff and I have to constantly like figure it out, you know, like it’s it’s crazy. but I would love to learn a little bit more about how you’re specifically using AI, in your work and whether or not you might be using one of these coworking apps, whether it’s like Claude Code or or Claude Cowork or Codex, and are you know, what do you find most useful about using the technology and then what are like some red lines for you in terms

of why what you would never use it for.

Sharon Goldman (34:50.264)

Sure. I am I would say that I use AI all day, every day, but it’s really as it really is as a coworker. It’s like as my editor, as my colleague. and that sounds I feel like that sounds kind of cliche in a way, but it really is true. Like I’m constantly and I’m I’m pretty much confined my efforts to using

Chatbots. I use Claude, I use Chat GPT, occasionally Gemini, but let’s say mostly Claude and Chat GPT. And sort of all day long, I’m sort of talking to it in a conversational way. If I’m working on a story, I’m doing research, I’m asking questions, I’m sort of getting feedback on thoughts. When I’m actually writing the story, I’m actually writing my story and then

A lot of times I’ll paste in a a sentence or something and I’ll be like, I feel like this could be better. Or I’m I’m there’s a word that I’m missing here and I can’t think of it. what do you or just what do you think of this? What do you think of this from a sentence structure standpoint? or or or how do you think this larger piece, how what do you think of my organization here? Could I

Should I be reorganizing it in some way? And this is just an ongoing conversation all day. that’s really different than using it for my writing or my reporting, which I just don’t do. I just don’t find it useful for that. I feel like it’s far more useful as an editor. You know, it’s and and that’s something that I mean at fortune I’m lucky enough to have amazing editors.

Pete Pachal (36:28.462)

Mm.

Pete Pachal (36:32.302)

Mm.

Sharon Goldman (36:38.668)

But even being able to hand in a story with an improved first draft from a copy editing standpoint is a step up. And previously, Adventure Beat, I didn’t have sort of editors on every one of my pieces. So, you know, it’s really helpful from that standpoint. As far as code Claude Code, Codex, these tools, I really haven’t gotten

used to using it in a regular way. I feel like partially that’s because as a journalist, I haven’t found the right use case for myself. I’m not, I’ve never been someone who has a lot of organizational workflows. You know, I’m pretty I’m pretty flexible and just kind of like as long as it’s in my Google Drive or my inbox, I can search for it. I I really haven’t found a use case that I personally find really useful for coding tools or development tools.

I also have found them to be a little bit hard for someone like me who isn’t naturally inclined to dig into them. I’m not someone who loves tools generally from you know pre-dating AI. So I feel like as they become more user-friendly for the average person, which I still consider myself to be as far as a tech tools user, I feel like maybe there will be some workflows that I find them useful for.

Pete Pachal (37:33.159)

sure.

Sharon Goldman (38:02.476)

But some of the things I use AI for are just so, you know, run of the mill. I mean, nothing has changed my life more than trans transcriptions, AI transcriptions. To be able to do that in real time on my interview saves me so much time. And that’s, you know, a pretty straightforward use case. But and being able to search those transcripts and being able to ask Claude or ChatGPT to find something in one of my transcripts.

Pete Pachal (38:12.942)

Sure. Yeah.

Pete Pachal (38:30.509)

Yeah.

Sharon Goldman (38:30.7)

These are the game changers for me as a reporter.

Pete Pachal (38:34.284)

Nice. I’d be interested in like circling back with you in like six months to a year and see where you’re at with coworking stuff, especially with w running your own operation, ’cause I’ve found them invaluable. Yeah. Anytime.

Sharon Goldman (38:42.604)

Yeah, I actually think that that yeah. Well, and I’d love to pick your brain brain on that. But yeah, I do think like running my own business, that’ll be a new opportunity for me to test certain things and see how they work for me and could I find them useful. so yeah. Well then I’ll have to take one of your courses then.

Pete Pachal (38:56.856)

Well, I think you’ll find them useful. Nice. wow. Hey, awesome. You heard it here, guys. sweet. So like but let’s look forward as we’re looking forward. I’m I’m curious about what you see as maybe the big stories in AI over the next year. We touched on the election a little bit, but that’s certainly gonna be something I feel like this is probably gonna be the AI election, at least maybe the maybe the first one that where it’s really a major issue. But what else what else are you seeing?

Sharon Goldman (39:11.552)

Mm.

Pete Pachal (39:25.774)

coming down the pike. what what do you think are gonna be the big stories? And also like what do you think will continue to be like the undercovered stories that that you think are the most important?

Sharon Goldman (39:36.889)

I think the AI data center boom story is going to continue to rise. Most notably the backlash. I think it’s growing. I think it hasn’t reached its peak. I still think it’s undercovered in some ways, like from some angles. For example, you know, of course I’ve been covering at Fortune, I’ve done a series of stories on different communities that have been affected by.

AI data center builds or proposals in their neighborhoods, in their backyards. But the backlash has become very activist. It’s become across you know, it doesn’t matter whether it’s in your backyard or not. It’s just people are generally opposed. And there’s been a tremendous rise in conspiracy theories, I’ve noticed. massive Facebook groups that are focused on the fact that they feel like

Data centers are surveillance centers, that they’re really about population control, that they, you know, emit radiation that can cause cancer. That, you know, I’ve e I’ve read some really strange things about people who think that they’re built on they’re being these mega AI data centers are being built on farms because they want to take away people’s ability to grow food. I mean, really out there kind of things. That kind of leads into another area that I’m concerned about, which is the potential for, you know, that backlash to turn violent. I do think that that’s a real issue. So that’s, you know, more things that I think could be reported. the build itself, I do find really interesting, like the different ways that data center companies are working to make those data centers more efficient, smaller, less, you know, impacting the environments.

There will continue to be more discoveries and improvements in that area. And I would like to report on that. I think the AI security issue is going to be huge. I’ll be going to the annual Black Hat Security Conference in August again, and DEF CON, which is the biggest annual hacker conference. And I think it’s going to be incredibly, you know, filled with news, which I’d like to cover because I think that’s

Sharon Goldman (42:02.976)

Cybersecurity is going to be a huge issue. and of course, you know, the the big IPOs are gonna be huge, you know, anthropic and open AI and SpaceX, you know, so from a business standpoint, you know, that’s just gonna be filling the headlines for for months to come. yeah.

Pete Pachal (42:05.976)

Totally. Yeah. With this whole mythos model, yeah.

Pete Pachal (42:12.683)

Mm.

Pete Pachal (42:26.85)

Nice. Well, going going a level up from the headlines, like thinking about the trends that all these things are feeding and and creating, looking forward, like what do you see? This is how I sort of wanna wanna at end most of my podcasts, like give me one thing to be concerned about in a broad sort of trend kind of way, and one thing to be hopeful about that from your perspective in in the AI world. Do it in whatever order you like.

Sharon Goldman (42:52.14)

I think that I think the biggest thing to be concerned about is how society is how how c AI is being communicated to society and how people are understanding it. I think the tremendous backlash to AI generally, like we’ve been talking about, is I mean, you can call it a communications issue, you can call it a technology issue, but whatever it is something is not getting through. This is not like previous technology evolutions where people saw some upsides. People saw upsides from the automobile, people saw upsides from the industrial revolution, from computers, from the internet, from social media. I feel like AI is having a harder time for all the reasons that we said. It’s it’s hard to communicate, it’s hard to explain. There’s

It’s so fast moving. There’s so much coming. And yet, is it any of it really life-changing? Plus, there are all these dangers that keep being talked about, as well as job loss. So I’m really concerned about how the general public is dealing with a societal shift that is so all-encompassing and fast and really hard to understand too. You know, these are people who are not tech forward, you know, for someone who’s not a techie.

on the other hand, I feel like that same thing, the idea of AI as this general technology is also what makes it hopeful because I do think that there’s another way to see it. It it it remains to be seen if job loss is really going to occur. I personally feel that there’s also a lot of jobs to be created. I think there’s a lot of areas that will, you know, maybe grow as a contrast to AI, you know.

in in real life experiences, for example. I do think there’s tremendous opportunity for AI to help with many issues in our global world, whether it’s healthcare or the environment. And there are also going to be so many technological improvements in terms of efficiency and you know, not being as impactful to the environment that I I do think there’s a tremendous

Sharon Goldman (45:16.212)

scope that we can look at for hope.

Pete Pachal (45:19.662)

Nice. Cool. We’ll leave it there. Sharon, thanks so much for dropping by. can’t wait to see what you do with ground level AI. And I can’t wait to have you back once once you’ve done some more incredible stories. Appreciate it.

Sharon Goldman (45:33.72)

Thank you so much, Pete. Thanks so much for having me.

The post AI’s reality check: Why Sharon Goldman is looking beyond the hype appeared first on The Media Copilot.

Tech companies are becoming Media companies. What happens next?

The Copilot — Fri, 12 Jun 2026 12:00:00 +0000

This episode is sponsored by: Adobe Acrobat

What happens when technology companies stop advertising in media and start owning it?

In this episode of The Media Copilot podcast, host Pete Pachal speaks with Jonathan Hunt, VP of Media & Content at HubSpot and Head of The Hustle, about the growing convergence of technology, media, and audience ownership.

As AI transforms how people discover information, search for answers, and build trust online, companies are rethinking their relationship with media. HubSpot has quietly assembled one of the most ambitious media portfolios in business, spanning newsletters, podcasts, YouTube channels, creator partnerships, and acquisitions including The Hustle, Mindstream, Starter Story, and Futurepedia.

Their conversation explores why audience ownership has become increasingly valuable in an AI-driven world, how brands are adapting to changing discovery patterns, and why trusted human voices may be more important than ever as generative AI floods the internet with content.

Listen or watch:

YouTube

Spotify

Apple Podcasts

Jonathan shares how HubSpot transformed from a software company into a media powerhouse, building a portfolio of trusted brands through acquisitions, creator partnerships, newsletters, podcasts, and AI-driven content strategies. He discusses the challenges of maintaining editorial independence while scaling audience growth in an increasingly competitive information ecosystem.

The discussion also dives into Answer Engine Optimization (AEO), AI’s impact on publishing, and the changing economics of audience development. As AI reshapes how people discover and consume information, Pete and Jonathan explore a critical question for the future of media: will audiences place their trust in publishers, creators, platforms, or the technology companies now operating at the intersection of all three?

Sponsor:

The new Adobe productivity agent orchestrates tools and models to generate images, text and rich content like presentations, podcasts and social posts, while also powering conversational PDF editing in Acrobat.

Link: Do that with Acrobat: AI-Powered PDF workspaces | Adobe Acrobat

Why this matters

The rise of AI is changing how audiences find information. Traditional search is giving way to AI-powered discovery, and companies are increasingly looking beyond advertising to build direct relationships with audiences.

As content becomes easier to produce, trust becomes more valuable. Organizations that own audiences, cultivate expertise, and build authentic relationships may be better positioned to compete in an environment where AI-generated content is abundant but human credibility remains scarce.

The discussion highlights a broader shift taking place across media and marketing: the growing realization that audience ownership may be just as important as product ownership.

What we cover

• Why AI may increase the value of human-created content rather than diminish it

• How HubSpot built a media network through acquisitions, launches, and creator partnerships

• The strategic thinking behind acquiring brands like The Hustle, Mindstream, and Starter Story

• Why audience ownership matters more than ever in a fragmented media environment

• How AI-powered discovery is changing traffic, engagement, and customer acquisition

• What Answer Engine Optimization (AEO) means for publishers and marketers

• Why visitors arriving from ChatGPT and other AI tools often convert at higher rates

• The growing influence of YouTube, LinkedIn, Reddit, and creator-driven media on AI search results

• OpenAI’s acquisition of TBPN and what it could signal about the future of technology-owned media

• How HubSpot balances editorial independence with corporate ownership

• The role of AI in content creation, production workflows, and operational efficiency

• Whether concerns about a “SaaS apocalypse” are reality or industry hype

• Why authentic creator partnerships outperform traditional influencer marketing

About the 👤 Guest

LinkedIn

HubSpot Author Profile

About the show: To explore more conversations like this and see what’s new, visit the Media Copilot website at mediacopilot.ai. You’ll find new episodes, expanded resources, and tools designed for journalists, communicators, and media leaders navigating the fast-changing world of AI. It’s the home base for everything Media Copilot and it’s just getting started.

Enjoyed this episode?

Produced by Pete Pachal and Executive Producer Michele Musso
Edited by the Musso Media Team

Music: “Favorite” by Alexander Nakarada, licensed under CC BY 4.0

The post Tech companies are becoming Media companies. What happens next? appeared first on The Media Copilot.

The AI industry has a Gen Z problem

Pete Pachal — Tue, 09 Jun 2026 12:00:00 +0000

Two years ago, if anyone had told me the most AI-hostile demographic in 2026 would turn out to be Gen Z, I would have laughed. The generation that grew up with screens in its hands seemed like the obvious early adopter, ready to use the tools to look more skilled, more productive, and more employable than everyone else.

Instead, May commencement season turned into an open revolt. At the University of Arizona, students booed Google chairman Eric Schmidt for pitching AI’s world-changing potential. Gloria Caulfield, VP of strategic alliances for the investment firm and real estate developer Tavistock, drew the same reaction at the University of Central Florida when she compared the rise of AI with the Industrial Revolution. At Middle Tennessee State University, students shouted down Big Machine Records CEO Scott Borchetta for the offense of saying the word out loud.

The numbers back up the booing. A recent Gallup poll measuring AI adoption and attitudes among Gen Zers found the share who say they’re excited about AI fell from 36% to 22% in a year. The share who say they feel anger toward AI climbed from 22% to 31%. Older cohorts are skeptical too, but a sentiment swing this sharp inside the youngest part of the workforce is something I’ve never seen for a new technology this early in its lifecycle.

That swing matters because the AI industry has a serious PR problem at the worst possible moment. Anti-AI sentiment is hardening as the midterms approach, and politicians are picking up data centers as a wedge issue, supporting efforts to halt or slow the build-out of the facilities that fuel AI with the computing power it needs to function. If capacity can’t keep up with demand, the cost of compute will keep rising, and that will put hard ceilings on what newsrooms, marketing teams, and comms shops can actually do with AI in the year ahead.

The infrastructure squeeze is already here

It’s already happening. Anyone who runs Claude as part of their daily workflow knows the rhythm of the outages, and Anthropic’s own status dashboard tells the story in red over the past 90 days. The Claude Code boom has driven demand through the roof in 2026, and the company is scrambling to keep up. Anthropic signed a deal to buy computing power from Elon Musk’s SpaceX, and at the same time it closed the loophole that let builders run third-party software on top of their Claude subscriptions. Some of those setups were burning thousands of dollars of compute against a $200-a-month Claude Max plan. Now those teams have to use Anthropic’s platform directly or move to pay-as-you-go.

The angry posts in response were predictable, but the more useful read on the change is that it forced builders to reckon with the actual cost of what they were running. The choices are probably familiar to anyone trying to budget AI spend: switch to a cheaper model, possibly an open-source one, rebuild on Anthropic’s own platform, or shut the project down.

This was always going to happen at some point. As demand grows, free-compute workarounds will keep closing. The industry’s argument is that the squeeze would hurt less if compute were cheap and plentiful, which is the case being made for trillion-dollar infrastructure projects like OpenAI’s Stargate. For AI to deliver on its promises, compute has to flow like water. That means more data centers, and more power plants behind them.

Which loops back to why Gen Z is angry in the first place. Environmental concerns are near the top of their list, and AI’s energy footprint has only gotten harder to ignore since I wrote about it months ago. This isn’t a fight confined to politicians and podcasters anymore. At the companies I advise on AI adoption, employee surveys keep surfacing the same worry, and in some cases it’s starting to shape whether teams use AI at all.

Governance is the new AI strategy

You can argue about whether the pollution and water concerns are overblown. The cost question is harder to wave off. The leaders running the most ambitious AI programs are past the era of handing every employee a ChatGPT seat. They want agentic workflows, automated processes, and rapid prototyping through vibe coding, and they may be telling their engineers to get obsessed with “tokenmaxxing.”

It’s unclear so far how data center politics will play out. What leaders can actually control right now is governance. That’s more than running training sessions on which model does what, although that matters. Real governance is the balance between experimentation and direction. People need room to invent their own workflows, and the organization needs a way to make sure the compute it’s paying for is being put to good use. That doesn’t just mean “keeping costs down” it means accepting that the bill will sometimes be high and being confident the outcome will be worth it.

Through my consulting work with media companies and PR agencies, I’ve watched this play out in practice. One agency I worked with piloted a vibe-coding tool. Usage spiked early as employees tested its limits, and several different teams ended up building near-duplicate prototypes. The thing that saved them was high-bandwidth communication. They ran regular workshops and project reviews, learned from their people, and steered the work as they went. They eventually homed in on the use cases that actually delivered, in their case automating media intelligence, and the experiment surfaced something unexpected. The original platform wasn’t the right one. The agency ended up adopting a different tool and sunsetting the one it started with.

That’s one of many examples, and the lesson behind all of them is the same. If AI agents are going to do real work for your team, they need to run compute-heavy jobs. Compute is going to stay expensive for a while. The way to avoid the kind of top-down restrictions that suffocate innovation is for leaders to define what success looks like up front, get their teams trained on the tools and models, and build systems that surface collaboration and catch waste before it compounds.

That is what good governance actually looks like. The political fight over AI and data centers isn’t going anywhere, but the companies leaning into AI can still find a way through. The goal is to work inside the real cost constraints while shielding the people doing the actual work from feeling them.

A version of this column appears in Fast Company.

The post The AI industry has a Gen Z problem appeared first on The Media Copilot.

Local news, AI, and the fight for accountability

The Copilot — Thu, 04 Jun 2026 04:34:55 +0000

This episode is sponsored by: Adobe Acrobat

What happens when artificial intelligence meets one of journalism’s most important missions: holding power accountable?

In this episode of The Media Copilot podcast, host Pete Pachal speaks with Rick Hirsch, director of the Collier Prize for State Government Accountability at the University of Florida and former Managing Editor of the Miami Herald. After more than four decades in journalism, Hirsch has witnessed nearly every major transformation in the media industry, from the rise of the internet and social media to today’s AI revolution.

Their conversation explores how AI is changing investigative and accountability journalism, not as a replacement for reporters, but as a powerful tool for uncovering stories, analyzing massive datasets, and helping newsrooms stretch increasingly limited resources.

“I think that’s what motivates most people who do this work. We have the ability to highlight wrongs and give people the information to try to right them.” — Rick Hirsch

Hirsch shares insights from a recent survey of journalists, discusses emerging AI-driven accountability tools being used by organizations like CalMatters, and explains why local government reporting may be one of the areas where AI can make the biggest positive impact.

At the same time, the conversation tackles difficult questions about trust, misinformation, newsroom economics, audience fragmentation, and whether journalism can sustain itself in an AI-mediated information ecosystem.

Listen or watch:

YouTube

Spotify

Apple Podcasts

Why this matters

While much of the AI conversation in media focuses on content generation, traffic disruption, and business models, accountability journalism presents a different challenge.

Investigative reporting relies on verification, judgment, sourcing, and public trust. AI can accelerate research and surface patterns that humans might miss, but it cannot replace the reporting instincts, ethical decision-making, and community engagement that make journalism valuable.

As local newsrooms continue to shrink and public trust remains under pressure, the future of accountability reporting may depend on how effectively journalists learn to use AI without sacrificing the standards that define their work.

Link: Do that with Acrobat: AI-Powered PDF workspaces | Adobe Acrobat

What we cover

Rick Hirsch’s 42-year career at the Miami Herald and his transition to the Collier Prize
Why accountability journalism remains essential to healthy communities
What journalists really think about AI according to recent newsroom surveys
How AI can help reporters analyze documents, legislative records, and public meeting transcripts
The innovative work behind CalMatters’ Digital Democracy project
How AI can help uncover story leads and reporting opportunities
Why shrinking local news coverage creates accountability gaps
The growing challenge of misinformation and declining trust in journalism
Whether AI can help combat accusations of media bias
Where the ethical red lines are for AI in investigative reporting
The future of subscriptions, memberships, and journalism business models
Why reporters still play an irreplaceable role in holding governments accountable

Key takeaways

AI is a reporting assistant, not a replacement

For accountability journalism, AI’s greatest value lies in helping reporters process information faster, identify patterns, and uncover potential leads. Human verification and editorial judgment remain essential.

Local journalism may benefit the most

As newsroom resources continue to decline, AI tools can help journalists monitor government meetings, legislative activity, and public records that would otherwise go uncovered.

Trust remains journalism’s biggest challenge

Technology alone cannot solve declining public trust. Transparency, documentation, and showing audiences how reporting is done may become even more important in the AI era.

The business model remains uncertain

The industry continues to grapple with how journalism will be funded as AI increasingly becomes an intermediary between audiences and information.

Accountability reporting still matters

Despite economic pressures and technological disruption, journalists remain motivated by the ability to expose wrongdoing, inform communities, and create meaningful change through their reporting.

About the show:

To explore more conversations like this and see what’s new, visit the Media Copilot website at mediacopilot.ai. You’ll find new episodes, expanded resources, and tools designed for journalists, communicators, and media leaders navigating the fast-changing world of AI. It’s the home base for everything Media Copilot and it’s just getting started.

Enjoyed this episode?

Produced by Pete Pachal and Executive Producer Michele Musso.

Edited by the Musso Media Team

Music: “Favorite” by Alexander Nakarada, licensed under CC BY 4.0

TRANSCRIPT :

Pete Pachal (00:30.414)
Hi, welcome to the Media Copilot. It's a podcast about how AI is changing media, news, and communication. I'm your host, Pete Paschal. I cover tech for a long time as a journalist, and now I talk with the media leaders, builders, and the creators trying to answer the question, how will we get information in the future, and how will that transform journalism and the business of media?

Quick housekeeping note if you're listening on Apple or Spotify, please leave a five-star review and maybe a nice comment And if you're watching on YouTube, please like the video and subscribe to the channel if you don't mind Those things really do help more people find the show My guest today is Rick Hirsch director of the call your prize for state government accountability at the University of Florida

Rick spent decades at the Miami Herald, including as managing editor, and he now works closely with the kind of journalism that's hardest to do and easiest to lose, reporting that holds state and local governments accountable to the people. That's where this AI conversation gets interesting, I think. In a lot of media conversations, AI gets talked about as a writing tool, or it's a traffic killer, or it's a business threat, but in investigative journalism, the question is a little bit more specific.

It's really about can AI help reporters make sense of all the documents and spot patterns and speed up their research and do all that without weakening the verification and the editorial judgment that makes the work trustworthy. So today I'm going to talk to Rick about the state of journalism, but especially investigative journalism and how AI is being used in that kind of work where certain red lines should be perhaps.

More importantly, maybe how the culture of newsrooms is changing as these tools become ever more present. Rick Hirsch, welcome to the Media Co-Pilot.

Rick Hirsch (02:25.2)
Thanks Pete, glad to be here with you.

Pete Pachal (02:27.744)
Awesome. So I gave kind of the short version there. Honestly, I look at your LinkedIn and it's a little bit like career goals if you're a journalist, right? Like you've done so much good work with the Herald and other things, but really give me like the quick rundown of your career and how you've spent it in accountability journalism, which is good term. don't hear it all that.

Rick Hirsch (02:49.532)
Sure, thanks. Well, 42 years at the Miami Herald started out covering government and politics. That's sort of my background and my passion and became an editor. Don't know why somebody thought I should be an editor, at least certainly didn't at the time. And in that role started out overseeing community based reporting in Miami, local governments, large and small.

and at a certain point began managing more people. Always kind of missed the story piece of it, but overseeing investigations at the Herald and we've had a pretty good run over the time that I've been there. And after doing that for 42 years and the last few years of both the first Trump administration and COVID and

In South Florida, the Parkland shooting, the Pulse shooting in Orlando, the collapse of the building in Surfside decided at a certain point that having my laptop open every night was something I had enough of.

Pete Pachal (04:05.998)
you

Rick Hirsch (04:06.556)
I had had a career recruiting reporters and for a period of time did some teaching at Florida International University and as a recruiter for the company that owned the Miami Herald, McClatchy, for all of its 30 newsrooms. And then this opportunity came along to oversee at the time a fairly new journalism prize focused on state government accountability at my alma mater, the University of Florida. And that seemed like a really important and

fun thing to do. And so that's what I've been doing for the last year and a half overseeing the Collier Prize.

Pete Pachal (04:43.084)
Nice. Yeah, tell us more about that. Like, why does it exist? How did it come to be? What drew you to it?

Rick Hirsch (04:48.24)
Well, of all things, there's an apartment developer in Gainesville, Florida, who has a real passion around journalism. His name is Nathan Collier. His great, great, great, great, great uncle. I'm not sure if I have all the greats right. Was the founder of Collier's Magazine in the 1880s. And he considers

investigative journalism and journalism in general, part of his family legacy. And so he funded the prize starting in 2018. It was just an annual prize. And then about two years ago, two plus, he gave the University of Florida an $8 million endowment to permanently endow the prize, hold an annual symposium on investigative reporting, focused on state government related institutions. And that's evolved into now we have a quarter

We do an annual survey of journalists on their attitudes and challenges. And of course, the big prize, the $25,000 prize, which most years is given out at the White House Correspondents Association dinner. This year, the award winners were there, but the prize didn't get given out for reasons that we don't need to get into in this.

Pete Pachal (05:59.607)
you

Pete Pachal (06:03.83)
Right, yeah, there are a lot of other sources on that one. No, we're not here to investigate that, but totally understandable. Yeah, that's great. I love that it's obviously not just an honor, but it has that financial part of it, finances are very challenged in journalism at the best of times. So that sounds really helpful.

So, okay, so I wanted to get a little bit into your career just because obviously it's very interesting, but also like you've clearly seen digital changes again and again in newsrooms and the media at large, know, the shift to just the shift to digital in general, the shift to social, mobile, all these things that have changed, not just how it's distributed, but you know, I'd argue also the practice of journalism on some levels. And now it seems like we're going into potentially the biggest change. I mean, I'll get your thoughts on it, but it's this AI moment must feel somewhat familiar to you, but also a bit different. What is your sort of, you know, zoomed out thousand foot view of AI and what it's doing to journalism?

Rick Hirsch (07:18.224)
Well, and I come at this both as a practitioner and a newsroom manager. And so I understand the concern reporters have. And I should say, I've been a media manager through the internet age, actually oversaw websites for the Miami Herald at a certain point. So.

We've gone through incredible loss in terms of resources in newsrooms. And there's understandable concern about the impact AI could have on that. My kind of perspective is the loss already happened from 2007 to last year.

Huge losses in resources in newsrooms. And there is some concern, rightfully, what AI can do, how it can be used to reduce resources in newsrooms. But I think a lot of journalists are sort of missing the power of AI in expanding their ability to cover things that they no longer can do because of the loss of resources.

So my background, my start was in covering small towns across South Florida. There was a time when the Miami Herald had 45 reporters covering local communities, local communities in Miami, and now they probably have five covering those local communities. But there is this potential with AI to get that view from 30,000 feet of what's going on in some of those communities and make smart choices about where to put the limited resources that newsrooms now have. So I think that's an exciting thing. I think that's a real opportunity. One of the things that we'll probably talk about a little bit, we did a survey of journalists and their concerns about AI and seeing it as both a threat.

Rick Hirsch (09:24.764)
...and a value, I think focusing on the threat for a lot of reporters is missing the point of how they can use this to increase their own value.

Pete Pachal (09:38.434)
Yeah, let's talk a little bit about that and some of the findings. So, you know, like I do feel like this is a rapidly evolving thing. Certainly my experience was that there was a lot of skepticism, certainly at the outset of the AI, I guess the current AI wave and right, probably rightly so, given the quality of the models and sort of that folks hadn't really figured out how to use them in the right way. Fast forward three years.

And now I feel like a good chunk of, a good chunk, not all of it, but a good chunk of it has kind of been worked out to some extent and guardrails are in place. in my limited experience, there's been at least some thawing of that skepticism. And there's a little bit more, whether it's excitement or just giving into the inevitability, I'm not sure. I'd love to hear your thoughts on it and maybe even think about...like first of all, like what the survey showed and also like, you see different textures in different parts of the newsroom?

Rick Hirsch (10:43.344)
Well, so the survey showed a majority of journalists seeing this as some potential threat, which is legit. At the same time, that same cohort, about 18 % of them were actively using AI in any kind of capacity. And so I think there's a lack of understanding of how it can advance the plot for you as a reporter.

how it can more quickly get you to a source you didn't know about, an expert on a topic, or a document that might take hours to find that you could find quickly, or ways to going through agendas or background material to find the thing that you really need. And of course, it requires humans.

to vet that information, make sure what you're getting from AI is accurate. But its ability, I think, to advance the plot has incredible value. in the area that I've been interested in, both state and local government, there are some newsrooms that are, and some resources that are really making incredible strides, I think, to improve how we do this work.

Pete Pachal (12:05.976)
Yeah, what are you thinking about specifically? Obviously, you're like accountability journalism. sort of say investigative, but I guess accountability is sort of more of an umbrella term because investigative always implies like giant feature, like your spotlight from the Boston Globe or something. And I think there's a lot of day to day work that's very important too, obviously. But are you thinking about specifically grounded tools like Notebook LM or even Google Pinpoint that
go through document troves or maybe even just homegrown stuff that does that kind of thing. And these tools that sort of not just do research but also help you process and analyze like just huge data sets.

Rick Hirsch (12:48.538)
Yeah, I would say all of the above. And there are some great examples. I think the best example at the state level is what CalMatters has developed with its digital democracy project. for those who aren't familiar, so the digital democracy tool that CalMatters developed ingests every video, every word spoken in the California legislature.

Pete Pachal (12:50.156)
Mm-hmm.

Rick Hirsch (13:17.852)
They pull in campaign contributions to every state legislator and using AI and analyzing that. It is a tool that both journalists and the public can use to try to find out what's being said and how legislators have voted on you choose the topic. And at the same time, it also creates what I would call a sort of story menu, potential story menu, it finds anomalies in how a legislator may have voted on issue A that is contrary to how they have typically voted on that issue in the past and gives a reporter an opportunity to look into whether there are specific campaign contributions on that. it generates kind of a digital tip sheet that CalMatters makes available to journalism organizations across California, not just their own, to pursue stories if they wish. And these are tips. They're maybe a little better, but maybe no better than the tips you get from the people who call you randomly when you're sitting in a newsroom and suggest you check out X, Y, or Z. They certainly need to be checked out.

But there are some really good leads that this system has created that have turned into good stories, really good stories, not just for CalMatters, but for TV affiliates in California, for other local newsrooms across the state. they just got a pretty hefty grant, I think $9 million from the Trust in America Institute.

I think it's $9 million over five years to try to develop a digital democracy tool to expand that to other states. And I think that has huge power. There used to be dozens and dozens of reporters who covered the state legislature in California. I'm in Florida, same is true. Now a fraction of that are there on a daily basis, but...

Rick Hirsch (15:35.066)
The potential of this giving access to both reporters across states and citizens groups to monitor what their state legislators are doing is really tremendous. So that's a really encouraging sign. There's another organization called Seagov, which just got a night grant. They're very active in the Bay Area that ingests video from local governments. So we're talking about the state level.

Pete Pachal (15:47.373)
Nice.

Pete Pachal (15:53.964)
yeah, I of this one.

Rick Hirsch (16:05.484)
C-gov is really focused more on smaller governments. not, I mean, they would certainly help the Boston Globe or the Minneapolis Star Tribune or the Miami Herald, but their real focus is on some of these nonprofit startups that are more hyperlocal and they're ingesting video from local government meetings and then...

using an internal AI tool. This is not something that they're not pulling in information from everything that chat GPT might be pulling in or whatever the noise that tends to lead to hallucinations when you're work with AI. They're just pulling that in from the actual government meetings and what's spoken and the agendas and et cetera. And creating

Pete Pachal (16:39.286)
Right.

Pete Pachal (16:55.491)
Go.

Rick Hirsch (16:56.976)
ways for journalists to search and find the most important thing that was said at a meeting or.

Pete Pachal (17:02.962)
Yeah, and video and audio also just kind of like a little bit more challenging in terms of those general rag systems, you know, they're gonna, which obviously are processing a lot more and want to just like summarize and or look for summaries. So yeah, it's great. You need you definitely need a targeted tool for those kind of things. And luckily, most local governments do throw a lot of that stuff up on YouTube. And it's just there. You just need the right tool to look at it.

All that sounds really great. like the though I've got to say the report does identify a lot of challenges in accountability journalism today, right? Beyond just resources, you know, going from like, you know, dozens of reporters to just a few. So I think it's sites, things like access problems are more prevalent today. That's interesting. And I'd love to hear about that. But, you know, there's also just the factor of disinformation and I, you know, all those terms, disinformation, misinformation, they get thrown around. But I think whether, whatever you, how to define them, I think it's very clear. There's just a lot more noise out there today. and then there's just the whole thing, like, is the public, do they even care? You know, like there's just, whether that's a symptom of just too much stuff out there, getting their attention, et cetera. so tell me about the challenges, but also.

Can AI, when used as a tool, help counter these particular challenges directly? And maybe I'm probably leaving some out, so you feel free to fill in the blanks.

Rick Hirsch (18:39.386)
Well, I mean, I think it can help. But I think that's where the humans come in. Because everything needs to be fact-checked. It doesn't matter where you get it from. I mean, even some of the tools we were talking about that are analyzing video and...

Pete Pachal (18:46.094)
Mm-hmm.

Rick Hirsch (19:01.23)
and sorting and categorizing it so it's easier to search, you still need to go to the video clip and make sure that the text, for example, that it generated is accurate, right? The artificial intelligence tools can't knock on doors and interview people, and they can't wait outside the mayor's office till he comes out or she when.

he or she doesn't return your phone calls. And so I think it's really important to remember that those things that journalists do still have to happen. mean, the issue of trust is a huge one. So there's all sorts of data that shows how...

people rely less and less on what used to be considered trusted sources for news. And I think anyone who's in the news business is faced with this reality that I think a lot of public officials from the top down will characterize any story that portrays them in a negative life as just being fake. It's become the go-to. It doesn't matter.

Pete Pachal (19:57.038)
Right.

Rick Hirsch (20:19.974)
how well documented it is, they will still at least initially say, that didn't happen. That's fake news. It's our enemies out there. So how that's combated.

Pete Pachal (20:28.75)
Mm.

Rick Hirsch (20:35.782)
to regain the trust that's lost. mean, part of it is show your work always. Being able to quickly, and AI can help do this, go to the video when you're writing a story about what was said in a public meeting. Being able to show the actual clip can really have impact. But it is.

Pete Pachal (21:04.59)
Mm.

Rick Hirsch (21:04.892)
Probably the biggest problem journalists face beyond resources is this lack of trust and lack of focus.

Pete Pachal (21:13.082)
And what do you think about the specific accusation of bias, which is prevalent in, it seems like in the world now. And is, I'm always curious, do you think AI has a role in either fighting that or disproving that, or even just giving maybe whether it's true or not, some sort of layer of perceived objectivity, you know, cause there's like sort of a... putting aside political bias, there's also kind of like an automation bias that people have, like they trust machines sometimes in weird ways. know, people will drive their cars into lakes just because like their navigation told them to.

So I don't know, I think about this sometimes and like, is this as AI becomes more of an intermediary between people and the information they got. Are there, there's obvious issues with that, are there upsides to that? I don't know, how do you just think about the issue of bias and...
know, real or perceived.

Rick Hirsch (22:13.222)
Well, I don't know how AI solves that as it becomes more and more efficient at creating things that are fake but are hard to detect as fake. I think we've all seen in the last year a big change from AI-generated videos where you'd notice the person only had four fingers.

There are fewer tells that something's manufactured than there used to be as it gets better and better. so, boy, how do we combat that? How do we make people trust? I can certainly imagine something crazy happening at a public meeting and having people in the meeting initially just say, well, that didn't happen. That was AI generated. How do you prove that it wasn't?

Pete Pachal (23:11.534)
Hmm.

Rick Hirsch (23:12.284)
I think that's a huge problem. mean, that's a huge problem. And what are the, we talk about what the guardrails are for journalists. And of course there's a lot of conversation over what the guardrails are for AI companies. And beyond that, if the company is developing AI can even control the machines they're creating.

Pete Pachal (23:35.183)
And what do you think about AI's role in other parts of the newsroom process? you know, is like, where are the red lines here? You know, is drafting ever acceptable? Does it depend on exactly what you're doing? Certainly, we as we've, it's kind of a given, obviously, everything needs human review, which you've already said, but

Are there things beyond just saying that? there just things that AI should never be a part of? Obviously, again, we've talked about the capabilities it simply can't do. Obviously, it can't win trust of a source or a reader. It can't pursue real journalism. think all that's a given. But then what else should it not do?

Rick Hirsch (24:21.87)
Well, I think when you're getting into investigative reporting, it can't write the story for you. It really can't. Can AI help you copy edit? Well, maybe, yeah. mean, it can certainly find spelling errors maybe better than a set of tired eyes.

in the 87th graph of 150 inch story. But the judgment it requires to put together an investigative story, think, I don't think AI should be doing that. I don't think reporters should be using that. I do think that there is value in using AI.

I mean, there are a of stories that I wrote as a young reporter that where I would rather have been doing something higher end. mean, AI can take the box score of a baseball game and write five paragraphs from it. And I think most sports writers would rather be doing a profile of player X than writing a bunch of those five inch stories. And I think you can take a real estate transaction and turn it into some text.

And most real estate reporters have their eyes on a bigger prize than writing that. So the degree to which it can free up reporters from the mundane to do the more important, I think is really valuable. But doing that more important really requires a reporter's judgment and an editor's review. in the case of an important...investigative story, a legal review.

Pete Pachal (26:21.208)
Totally. And if AI is also present in the ecosystem and you have these things where you spend a ton of time on an investigation and it's out there, and then some AI portal just kind of summarizes it. it's like, is that, because this is the thing that I feel like obviously is a big center of tension between the media and AI companies. I'm more curious about the reporter, editor, investigative level.

You want your stuff to be out there and you want people to access it. But at the same time, you want to make sure that you have, you know, both proper credit and the means to keep doing these kinds of things. So how do you, well, not just how do you feel about, like, is the right sort of balance here in terms of like using a, thinking of AI as a distribution platform for this kind of work.

Rick Hirsch (27:26.586)
It's really tricky. It's really tricky. So if high quality, vetted, investigative work is not available for the large language models to access, and there are lawsuits and negotiations and all sorts of things ongoing between news organizations and AI companies about how that ultimately should work, right?

But if that information is not available, but what is, is disinformation and members of the public are turning to AI to do their homework and that's not available to them, then what are they getting? I mean, that's one side of the argument. On the other side, news organizations, reporters need to be paid a living wage to do that quality work. And so if...

No one is reading their stories on their platforms or subscribing because they can get it all for free through an AI subscription or through search, then where are they? I'm going to just, not to cop out, but above my pay grade to figure that one out. mean, it's pre-AI.

When the entire issue was search, I spent a lot of time learning, coaching, training journalists to be really smart about using SEO. So audiences that really had no loyalty to a certain publication.

Pete Pachal (29:05.758)
Mm. Me too.

Rick Hirsch (29:15.628)
we're likely to find the news, the story by that publication and read it. And that's how, you know, the pennies went into the pocket of that news organization. And of course, AI is changing that. So I wish I, if I had that answer, I would be doing something else right now.

Pete Pachal (29:24.75)
Hmm.

Pete Pachal (29:32.75)
Hmm.

Pete Pachal (29:36.845)
Yeah.

Yeah. And so it's, it's one of these tricky things, which is interesting. I'm glad you brought up search and SEO because I always look at incentives, right? So it's like the search and social era had certain incentive. let's do keywords. Let's get on page one of Google. And it taking to an extreme that turned into these sort of like in the social case, these bite-sized provocative stories and culture reporting. And then in the search case, it's like, let's litter it with keywords. In fact, let's write articles just to do keywords like that.

So those incentives could be easily perverted and they were in a lot of places. AI has different incentives. I don't know if they're better. I'm sure they're different, but they are different. journalists are always kind of aware of these incentives on some level. Like you say, you go, I had to coach people on how search worked. And I've done that a lot myself because you got to be in the game. Like that's the thing I would tell people, right? It'd be like, look, it's not that...

you're writing for machines, but the machine is the intermediary now. And if you're not doing, taking care to be present in search, you're just not reaching anyone. And I feel like that's going to be the case with AI sooner than later, right? And so, you know, putting even aside the compensation question, we can talk a little bit about that, because actually I'm interested in if you're thinking like, if that isn't dealt with.

Well, actually, let's attack that head on. if that isn't dealt with, which is to say that you're not getting paid by the AI companies for this, because we're not sure where that's going to land. But to some extent, it's happening for a lot of publications. It doesn't seem like it's going to be an option. Does that necessitate moving toward a model that's more like memberships, events, like these other ways of monetizing content that aren't just like putting ads next to it the traffic just won't be there?

Pete Pachal (31:36.234)
If that is the case, could it be spun as better or adapted into something better? mean, you think about if you have this maybe smaller audience, but they're all members, there's more of a stronger engagement there. I don't know, where do you land on this?

Rick Hirsch (31:55.526)
Well, I don't want to be gloom and doom-y about it, but maybe this will be a little bit. So I think the subscription model is really important. And loyalty to a publication is really important. However, the challenge in that is the expectation of your subscribers. And we are in an era where many people want to read information that reinforces what they already think. And so if most of your subscribers lean right or lean left, you get blowback when you're publishing stories that are contrary to where they lean.

And the more dependent you are on that subscriber base, the more of a potential problem that becomes. So that's always concerned me about the subscription model. What I always liked about the subscription model is it's not geared toward clickbait. It's not geared towards, in Miami, the alligator that falls into a boat and everybody scrambles. Not that I'm against the story about the alligator in the boat, but that's not accountability journalism, right? That's just, you know, viral stuff. Right. So so the deep dive, I mean, the stories we know are most likely to cause people to subscribe are investigative stories, enterprise stories, stories you can't read somewhere else. And so, you know, I do think that's both a journalist, a journalistic good

Pete Pachal (33:26.944)
Exactly, you're getting addicted to those traffic spikes at that point. Yeah.

Rick Hirsch (33:52.764)
and a business strategy good, but I worry a little bit about how you might be painting yourself into a corner with a specific audience.

Pete Pachal (34:07.928)
So as we sort of wrap things up here, I'd love to try to unpack a little bit about some of the stuff on the positive side, getting away from the doomerism. Like the report did say that journalists remain committed to accountability despite all the pressures they're under, despite being research, you know, the resource strapping, the sort of difficulties that we've just talked about. What do you think is keeping that commitment alive?

Rick Hirsch (34:34.556)
Well, I think we still see how the reporting can make a difference. And one of the things that we do in the monthly newsletter that we published with Call Your Prize, we have a section every month where we list like a dozen to 15 stories across the country, accountability stories that have made a difference. And they really move the needle. They change laws. And that's why...

I think most of us got into this business. I'm fortunate enough, I spent my whole career at the Miami Herald and I grew up in Miami and being able to have an impact on the community I live in. I mean, some people do that through in civic life, they run for office or they work in local government or they work for a nonprofit. I was able to do that in journalism and I think that's what motivates most people.

who do this work. It's certainly not, I mean, we're never gonna get rich doing it. When I was in journalism school, my grandmother said, why don't you get into a profession? Anyway, so I think that's always been a motivator. It's still a motivator. It still matters. I think we have the ability to highlight wrongs and give people the information to try to write them. And that's the driving force behind.

Pete Pachal (35:33.196)
Thank

Rick Hirsch (35:56.508)
most journalists, how we can convince people that that's what drives us is part of our challenge going forward. But that's a big deal. And then

I'm encouraged by some of the tools that are being developed to use AI so that we can better do that, not just at the level where the New York Times is operating, but where some hyperlocal sites are trying to cover communities and figure out why they're going to build a 10-story building where that gas station used to be in the corner.

Pete Pachal (36:32.834)
Nice. You're kind of, think, partially answering my last question, which is going to be, it's always the same. It's looking out into the future, projecting out and thinking about the AI mediation of our media ecosystem and the practice of journalism. What is one thing that you're really concerned about and another thing that you're hopeful about?

Rick Hirsch (36:58.62)
So I'll the concerned part first. I'm concerned continually about the business model. And I think this is one more challenge. think journalism organizations have always been

three, four, or five steps behind technology in figuring out how to adapt what we do to the changes that are taking place. And the speed with which AI is evolving, think, is making the ground that's being lost is being lost faster. But I am optimistic about the tools it can provide.

to cover areas that have been lost over the last 20 years in terms of local government. And I'm hopeful that good, smart journalists will, while they need to protect their personal interests, will invest the time and effort to figure out how to make AI work for them in the work they need to do in covering communities.

Pete Pachal (38:09.89)
Nice. We'll leave it there. Hirsch, thanks for coming by the Media Copilot and sharing your thoughts. Cheers.

Rick Hirsch (38:15.324)
Pete enjoyed it.

The post Local news, AI, and the fight for accountability appeared first on The Media Copilot.

The click is dying but the citation just got more valuable

Pete Pachal — Tue, 02 Jun 2026 12:00:00 +0000

Google is not a company anyone expected to root for in the AI era. The early Bard demos were rough, Perplexity and ChatGPT were peeling away curious users, and antitrust regulators were closing in. The narrative a year ago was that the search giant had finally been outmaneuvered.

Now that looks dead wrong. Google is in a much stronger position today. Not because of it’s just coming off a prolific I/O developer conference, and not because it suddenly has the best model or the most capable AI ecosystem. Those titles get passed around the big labs every few months in any case. The reason is simpler: the money is still coming in.

The business is adapting

The Q1 numbers tell the story. Alphabet’s Google Services revenue was up 16% to $89.6 billion. Google Search and “Other” revenue was up 19%. The data is a strong indicator that the supposed AI disruption to its search products hasn’t dented the ad machine. If anything, it has fed it.

That confidence showed at I/O. Google announced many new AI products, but one of the most notable ones to the media industry was a set of new ad formats. Conversational Discovery ads are built on the fly to fit naturally into the answer to the person’s query, appearing as a “sponsored” section. Highlighted Ads and AI-powered Shopping Ads work similarly inside general product category queries. And then there are Business Agents for Leads, tailored versions of Gemini that live inside the ad itself.

These formats are still in testing, but the direction is obvious. Google is getting more sophisticated about how it monetizes AI experiences. A few months ago, the company stated it had no plans to sell ads in Gemini, a line executives floated in response to ChatGPT ads. Technically that line is still operative; Google can still say that the Gemini chatbot is not becoming an ad product. But that distinction feels less meaningful now that so many Gemini-powered AI experiences across Search are being commercialized.

Here is the part publishers should sit with. All those AI-powered ads appear within or next to an answer. That answer is built, in large part, from the work of media publishers. In the old system, Google sold ads next to results, and those ads benefitted from the close proximity to links from trusted media sources. Search the best SUVs and you may see ads for Toyota or Hyundai before you see a link to Car and Driver.

Now the information, built in part from the publisher’s content, is right there on the result. The user gets the info, the AI-powered ad provides a path to transact, and everything is handled without any need for them to ever leave Google. The shift is fundamental. Instead of monetizing the path to information, Google is now monetizing the information experience itself.

Publishers, of course, get cut out of that bargain. In many cases, their content was the raw material that informed the answer. Early in the AI search era, Google’s pitch to publishers was that AI-referred traffic was higher quality, more likely to engage and transact. That was, broadly, true. But why would users engage on a publisher site when Google is providing the means to do that before they ever arrive? The new ad formats are an acceleration of a trend that was already bad for publishers.

Trust is the variable everyone is missing

And yet. Users don’t care about business models. Whether they have an inclination to buy something or engage depends not just on the content of the answer but on how much they trust it. That is where the calculus gets interesting for publishers. A study published in Nature described trust in AI as dynamic and context-dependent. In other words, it changes depending on the nature of the AI experience and over time. A separate study by the Reuters Institute found users had moderate trust in AI answers, but they also value their speed and aggregation. Translation. Utility is high. Trust is conditional.

One of the most important assets any media brand has is the trust it cultivates over time. Imagine two AI answers about the same product. One built from social posts, blogs, Reddit threads, and online forums. The other built from articles on Consumer Reports, the Wirecutter, Time, and CNET. The user doesn’t need to know the methodology to feel the difference. Which one sounds more trustworthy?

Citations, in other words, are not decoration. People will be more inclined to trust answers created from brands that they’re familiar with. Hard data on AI ad performance is thin, but the entire media ad model is founded on this idea. An ad doesn’t just benefit from being present on a platform. It benefits from being associated with that platform’s brand. Ads inherit context. They always have.

Google has not, to date, been especially responsive to what publishers want. But Google does need advertisers to believe AI search ads work. That need is the leverage. If advertisers see better performance when ads appear beside credible, well-sourced answers, they will care about the quality of those answers. Once advertisers care, Google has to care. That could create pressure on Google to maintain a healthier source ecosystem.

What that pressure looks like is the open question. It may not look like simple licensing deals. It could involve clearer traffic paths, richer citation treatment, new publisher products, commercial partnerships, or advertiser demand for premium source environments inside AI search results. Each of those is a different commercial conversation publishers should be having now, not later.

The click fades but the value doesn’t

Review sites are the clearest example because the transaction path is obvious. If someone asks for the best dishwasher, the AI answer can cite reviews and then push the user toward purchase. But the same logic extends well past commerce. A health answer, a travel plan, or even a summary of a political issue all depend on source trust. Even when there’s no immediate checkout, the user’s confidence in the answer shapes what they believe and what they do next.

For publishers, the warning is straightforward. Google’s new push into AI ad experiences could further weaken traditional publisher revenue streams, especially traffic-based display, affiliate, and search-driven monetization. For practitioners trying to think a step ahead, there is another side to the equation. If AI answers need credibility to be useful, then credible media still has value. That value will not always show up on a referral chart. But it will still shape whether users trust the answer enough to act on it.

A version of this column appears in Fast Company.

The post The click is dying but the citation just got more valuable appeared first on The Media Copilot.

The Scraper Economy is already here. Publishers just aren’t getting paid.

The Copilot — Thu, 28 May 2026 05:01:17 +0000

This episode is sponsored by: Adobe Acrobat

On this episode of The Media Copilot podcast, host Pete Pachal sits down with Jonathan Woahn to zero in on a part of the AI content ecosystem that’s just out of sight.

The conversation explores the fast-growing “scraper economy,” where data brokers, indexing companies, and AI infrastructure providers are quietly monetizing access to the web at massive scale while traditional publishers struggle to establish sustainable licensing models. With this gray market of internet data growing, how can publishers both protect their content and take advantage of the now billion-dollar demand for it?

Pete and Jonathan also explore:
• Why the social contract between Google and publishers has fundamentally changed
• The rise of ethically sourced data and whether AI companies will eventually care where content comes from
• Why inference markets may become far more valuable than model training
• How publishers should think about MCPs, AI infrastructure, and product strategy
• Whether a legitimate marketplace for AI content licensing can actually emerge before scraper economics dominate the ecosystem

Along the way, Jonathan shares how his company, Cashmere, is helping publishers structure, license, and deploy content for AI systems while quietly brokering relationships between content owners and companies looking for legal, high quality access to trusted information.

Listen or watch:

YouTube

Spotify

Apple Podcasts

Why this matters:

As generative AI continues reshaping how audiences consume information, the future of publishing may depend on whether media companies can establish sustainable economic models around their content before gray-market scraping ecosystems become the default infrastructure layer of the internet.

This conversation goes beyond AI hype and digs into the economics, legal gray areas, and technical realities quietly redefining the relationship between publishers, platforms, and information itself.

About the 👤 Guest

Linkedin: jonathanwoahn

Website: cashmere.io

Linkedin: company/cashmereio

Manifesto: cashmere.io/manifesto

Link: Do that with Acrobat: AI-Powered PDF workspaces | Adobe Acrobat

Enjoyed this episode?

Produced by Pete Pachal and Executive Producer Michele Musso
Edited by the Musso Media Team

Music: “Favorite” by Alexander Nakarada, licensed under CC BY 4.0

TRANSCRIPT

Pete Pachal (00:20.341)

Hi, welcome to the Media Copilot. It’s a podcast about how AI is changing media, news, and communication. I’m your host, Pete Pachal. I cover tech for a long time as a journalist, and now I talk with the media leaders, the builders, and the creators, all trying to answer the question, how will we get information in the future, and how will that transform journalism and the business of media? Quick note.

If you’re listening on Apple or Spotify, please leave a five-star review, maybe a nice comment. And if you’re watching on YouTube, please like the video and subscribe to the channel if you don’t mind. Those things really do help more people find the show. This week, we’re talking about one of the biggest unresolved questions in media and AI, which is if AI systems are going to use publisher content to answer people’s questions, what does a real marketplace for that content look like?

Because the uncomfortable answer may be that the marketplace already exists. It just doesn’t include publishers. A growing network of scraping companies and data brokers and AI infrastructure players is helping companies access the web at scale. And some of that is powering AI search, rag systems, and even enterprise research tools. Now publishers are trying to stop some of it, maybe use some of it, and certainly get paid for more of it.

My guest this week is Jonathan Wolin, co-founder and CXO of Cashmere. Cashmere is a company building infrastructure for premium publishers to manage and license and protect and monetize their content in AI systems. Cashmere has worked around deals involving publishers and content providers like Wiley and AI platforms like Perplexity. And Jonathan’s been thinking deeply about this and the difference between a scraper-driven market and a legitimate content marketplace.

So we’re going to talk about this. We’re going to talk about the scraper economy, why publishers may be losing, what little leverage they have, what it could take to build a cleaner licensed market, and whether the future of media depends less on blocking bots or more on building just good pipes. Jonathan Woen, welcome to the Media Copilot.

Jonathan Woahn (02:31.714)

Pete, it’s great to be here, thank you.

Pete Pachal (02:34.389)

So before we get into all that big, big Haiti topics I was just describing there, I’d love to learn just like a little bit more about you and your background and how you came to be connected to this, the business of connections, which is connecting like publishers to good marketplaces. So tell us a little bit about your background.

Jonathan Woahn (02:38.434)

Yes.

Jonathan Woahn (02:55.436)

Yeah, thanks. I’m excited to be here, Pete. So my background, I’m a serial entrepreneur. This is the fifth company that I’ve actually started or has been early member of. And in the previous company that I was at, it was called Book Club. And what we were doing at Book Club is we were working with publishers and authors to create professional development programs. And we started out doing these bespoke manually. It was very time intensive. It was very expensive.

And we, when AI and chat GPT came around and we saw, we could actually use AI to create very custom and very bespoke professional development programs. And so we started trying to do that and we ran into two big problems. found, first of all, the process of actually licensing the content was extremely time consuming and difficult. And then the second was even once we were able to license it, the structure of the, the, the data that we got itself.

was very difficult for AI to work with. the anecdote that I use with people is, you know, we would create, um, you know, a program around seven habits of highly effective people. And early days, the AI would create the 26 habits of highly effective people. Like it just didn’t know how to like, kind of pull the information out of the content. Right. And so we just looked at this and said, you know, AI is not going anywhere. Premium content publishers need better rails to be able to interface and interact with AI and

Pete Pachal (03:57.653)

Hmm.

Jonathan Woahn (04:24.79)

We’ve had to build a lot of technology to understand how to make this work and how to facilitate it and how to get AI to work well with it. And so we ended up launching Cashmere to be that infrastructure to help support publishers and building, making it so their content can be used with AI.

Pete Pachal (04:41.141)

Yeah, like what you miss mentioned there really resonates. It’s like the frustrating thing when you get into these AI systems and you expect it to do something semi-deterministic, like surely you can adhere to a character count. And in the same way, like surely you know what the number seven is. And it kind of doesn’t, know, like it kind of does and kind of doesn’t. Now, to be fair, I think we were both, but we’re both kind of talking about AI circa 2024, 2023 probably.

Jonathan Woahn (05:08.238)

Yeah. 2022. Much better. 100%.

Pete Pachal (05:08.789)

The systems that we use today are better at this. But still, it really hammers home that the AI interpretation alone is not the thing you should be 100 % reliant on, in terms of cleaning data and putting guardrails, whether they are within prompts or in the systems itself.

just has to be a part of this process or it’s just not gonna be reliable.

Jonathan Woahn (05:38.382)

Correct. Yeah. Yeah. Especially as the publisher, like if you’re wanting the AI to represent the content in the way that you want it to be represented. Definitely. Yeah. It’s still.

Pete Pachal (05:49.065)

You got to have a voice here and that voice requires some knowledge about your data and treating yourself like a data company in many ways,

Jonathan Woahn (05:56.578)

Yep, 100%. Yep, that’s exactly right.

Pete Pachal (05:59.605)

Cool. So honestly, I’d love to just get right into it. Like the serious stuff I was talking about at the beginning. like, I know, you I, you, you’re on LinkedIn and you, you’re writing about, things going on and there’s some good research out there from Matthew Goldstein and, you were writing about how the AI content marketplace is like already there, but just like the publishers are not, are barely a part of it. Like, it’s not like they’re not there, but it’s like something like a 1.6 billion. I forget what the exact numbers are.

versus something that is like less than a 10th of that size in terms of the actual licensing money sort of being exchanged. So it’s like 14 to one, I think was the ratio you mentioned in terms of like the actual, there was the size of the market versus like the cuts publishers are actually getting in this market. tell us like what’s going on. How did this happen? How did we get here where there’s like just mass scraping of content that’s being sold at scale and publishers are just like, I wouldn’t even say they’re like,

Jonathan Woahn (06:41.474)

Yeah.

Pete Pachal (06:59.017)

barely keeping up, they’re like drowning in this.

Jonathan Woahn (07:01.528)

Totally. And it’s interesting, since then, the number is actually even worse. It’s actually 20 to 1, based off some more recent information that we’ve looked at. I mean, look, we’ve all grown up in this world that Google has kind of curated for us for the last 25, 30 years, where it’s like we’ve been like, we’ve reached some degree of equilibrium where

Pete Pachal (07:08.02)

Wow.

Jonathan Woahn (07:27.672)

You know, we’ve been comfortable with playing Google’s game on SEO and trying to figure out how to get content, you know, how to get it ranked and how to get it discovered. And one of the things that has happened with the advent of AI has been the democratization of search. And, and so like, what I mean by that is, you know, I mean, Google is the default place that we would all go to search and find, look for information. mean, it was just kind of like the first stop. And now what has happened is.

And now we’ve got chat GPT, we’ve got Gemini, we’ve got copilot, we’ve got Claude, we’ve got perplexity, we’ve got and, and, and, and, and, and some of these platforms are building their own indexes and they’re scraping the internet on their own and building their own index so that they can serve their own platform. But what is happening more and more is that, you know, as you and I are standing up, you know, Claude copilot or standing up our own agents to help with our own research or things that we’re doing on our own machines, like

They need access to content. They need access to the internet. And so what has happened as a result of this is, you you used to be able to use Google’s API for search, Bing’s API for search. have both since to my knowledge, shut both of them down. remember when Bing did it two years ago, it was like this big deal. was like, they saw that this competitive asset that they had built this internet scale index. like, we’re not going to make this available for everyone. We need to use this internally.

Pete Pachal (08:54.421)

Hmm.

Jonathan Woahn (08:54.53)

But then what that created was this huge vacuum for a lot of other players to come in and to start scraping the internet and to start building their own indices and to start selling, you know, the work that they’re doing. Cause I mean, candidly building a great crawler and building a great index, it’s not an easy problem to solve. And it’s not a cheap problem to solve. It’s, it’s very technically complicated and it’s very expensive. and so there are definitely opportunities for like economies of scale to come in and for some of these.

Pete Pachal (09:11.912)

Hmm.

Pete Pachal (09:17.012)

Hmm.

Jonathan Woahn (09:23.534)

crawling and scraping platforms to be able to provide solutions. And agents need access to that internet, that content. And so what’s happening now is you’re starting to see a lot of these guys that are popping up and selling content. And there’s like this race to the bottom on pricing because they’re trying to figure out how do we get, you know, how do we get as many agents using our search as possible? But the thing that they are not doing is they are not licensing that content from the publishers that are scraping it from, and they are not sharing revenue with anybody.

Pete Pachal (09:53.459)

Yeah, you’re really zeroing in on like, kind of like the sore spot here. It’s really more of a vacuum of, I don’t know if it’s law or best practices or a few other things, but the way the internet evolved, this indexing, this active indexing was treated as, you know, very benign in the sense that you have an index and that’s just going to help people find your content online. And that was all anyone ever thought that was ever going to be used for. And now

Jonathan Woahn (09:53.518)

Pete Pachal (10:20.741)

you know, obviously that still is part of it, but now with this content layer, this interpretive layer of AI put on top of it, that is owned and operated by these players that aren’t publishers, you know, it’s suddenly that bargain is gone, right? But the infrastructure still exists and it’s sort of like treated as almost like a given that, yeah, indexing, let’s just do indexing, but it’s less about what…

that the infrastructure and more about quote unquote the outputs, whereas in the old days it was like links and now it’s actual content. And this is where like obviously where the contention is and nothing’s really been decided it feels like as much as what you can and can’t do. So everyone just defaults to whatever they can do. There’s no should.

Jonathan Woahn (10:59.618)

Yeah.

Jonathan Woahn (11:11.406)

Yeah. Yeah. I mean, there was a, you know, there was this, you know, social contract between Google and all every website where it was like, I’m going to scrape your content and I’m going to show blue links and I’m going to redirect traffic. We’re going to get you, we’re going to get people to your site. Right. And these days, I mean, there is, there is no contract. A lot of these scrapers are just scraping the content and they are serving it up directly to the agents. And.

They may include the source link, they might not, but there’s nothing that effectively demands that the agentic interface or where it’s using or referencing any of that content for inference, there’s nothing that says that they have to point back to the source or that they have to redirect people. So a lot of that social contract has just been totally upended through a lot of the way that the agents are accessing these indexes right now.

Pete Pachal (12:07.679)

So I know the report that we’re talking about from Matthew Goldstein, is great. It’s been sort of going around. But it makes a distinction between training and grounding. Does that distinction matter that much to publishers?

Jonathan Woahn (12:22.408)

well, yeah, let’s, let’s yeah, go ahead. ahead. Yes. the shorter answer, the shorter answer is yes. Publishers should absolutely care about the difference between inference and grounding. and let me talk about, you know, let me define that. So, you know, or, sorry, I inference and grounding. mean, inference and training or grounding and training are kind of like the two people familiar with those.

Pete Pachal (12:25.459)

I guess we should define those terms.

No, you go ahead. You’re the expert,

Pete Pachal (12:50.057)

Right, sure.

Jonathan Woahn (12:52.056)

So training, the analogy that I’ve used in the past is like training is like as a student, you’re going to school, you’re learning your field of trade, an engineer, right? You go, you’re learning how the equations have and how they work, you’re paying to be there. But at some point, you know, and that’s like training of the model. You’re given all the information and learns how to do what it’s gonna do, right?

Inference is like, and grounding is like when you’re actually applying the skills. So now as an engineer, you graduate and then you go out and it’s like, okay, they’re going to hire me to build this machine. Now you’re actually applying what you’ve learned. And that’s like what inference does with these models, right? It’s like, now we’ve taken all the things that they’ve learned and like using it to generate output, output tokens back to, you know, respond to whatever the user request is. And in, in the world of publishing,

there are a lot of reasons why we need to understand the difference between these two and why it’s really important monetarily to grok the difference. On the training side, are the majority of, Rob Kelly has tracked a lot of this stuff around the training deals that have been taking place and the different deals between publishers. And a lot of the announced deals have largely been for training deals. And so,

From a publisher perspective, it’s pretty easy because you just package your stuff, send it over and you’re getting paid, right? Like you just align on like what those terms are and now you can start making money from training. and you know, the big question around training is like, you know, where does this land from a copyright perspective? Like, is this a transformative, is it something that’s going to be permitted or not? Right. And like the courts are starting to get some kind of clarity around that. The one that is very not clear right now is inference.

And my perspective on this, this is one of the things I was talking with some publishers about yesterday, was it’s like, whether training comes down on the side of copyright infringement or not, I don’t know that it totally matters a lot just because the financial opportunity around inference is so much greater than I believe that the opportunity around training is.

Pete Pachal (15:06.729)

Yeah, exactly. again, just to redefine and so that everyone knows, like inference is like accessing stuff in the real time stuff. Cause training takes a long time. takes, it’s very compute intensive. I don’t know. It’s like, I don’t know, every few months or so the models are retrained or something, maybe even longer than that. Obviously it’s still an important issue, but it’s like accessing like the news that happened today or even an hour ago.

Jonathan Woahn (15:28.61)

Yes.

Pete Pachal (15:30.537)

That’s the opportunity because people want that information. mean, you know, there’s a whole, it’s called the media industry. It’s based around this. You need current information often up to the second, if you’re talking about market movements. Yeah.

Jonathan Woahn (15:42.2)

That’s exactly right. And you can’t train it on that stuff. It has to happen all at runtime. It has to be at grounding. And Pete, it’s beyond just the currency of this, right? But there’s also just the historic record. And so for media publishers, it is going to be about what are the breaking stories, what’s happening right now? And for, say, an academic publisher, the academic record is

Pete Pachal (15:55.797)

Hmm.

Jonathan Woahn (16:07.854)

perpetually changing and we’re learning things that change over time. so publishers need to have the ability to say, here’s what I want available. Here’s what I want discoverable. Here’s what I want in the public record versus like, this is no longer relevant. We need to be able to retract this. We need to be able to pull this out. And you cannot do that from a training perspective. You sell your content, it’s baked into the soup, baked into the cake, right?

Pete Pachal (16:25.333)

Mm-hmm.

Pete Pachal (16:30.085)

All right. Snippets don’t matter so much on the training. this is really good point. I think early on, there was too much focus on the training because there just wasn’t a lot of inference, at least in terms of outputs. Now there’s a lot in terms of all the big AIs have some kind of search connected to them. And there’s a lot of third party systems, legit and shady, shall we say, involved.

Jonathan Woahn (16:32.897)

Right.

Pete Pachal (16:58.549)

So how much has that, damage is probably overstating it, but is there’s kind of a bit of a distraction on training in that, not that that’s been resolved, but is its relevance to what’s actually going to be a sustainable future seems minimal, is what we’re talking about. is that causing a sort of an education problem among publishers almost?

Jonathan Woahn (17:23.63)

I do think so. The dollars, I was just in with a publisher earlier this week and they were just talking about the impact of some of the training deals they’ve done on their budget for the year. And the training deals are some of the things that have kind of helped to kind of close the gap on the work that they’re doing. And so it’s like, they look at these and see like there is real dollars here and it is making a real impact to their top and to their bottom line.

Pete Pachal (17:44.66)

Interesting.

Jonathan Woahn (17:54.146)

but it is episodic, it’s not predictable, the training market doesn’t have the signals of a scalable, sustainable market, right? It’s harder to see, but I think that has, and it does distract from the inference-based opportunities. And candidly, I think a lot of these scraper platforms are distra…

are part of the reason why it’s getting distracted is because they’re not seeing the dollars because those dollars are being diverted to other companies that are not the publishers themselves.

Pete Pachal (18:28.883)

Is this a part of like why a lot of, feels like the deals have kind of dried up. Like there was a bunch for a while there and a good deal of them were open AI, although some of the other ones weren’t necessarily as public. So again, I guess I’m asking you in terms of what you’re hearing, because you probably talking to tons of people all the time. But I do feel like the deals have gotten fewer and farther between. And I think maybe that’s partly a,

demand issue, if you know what I mean. So in the sense of like, um, we’ve been talking about these sort of scraper companies that have risen up in the last year and a half, two years. Um, and they’re, they’re in this legal gray area, certainly, but it’s rather than making a deal with a publisher, like maybe you could just go over there and it’s, you know, to this sort of gray market and get your stuff. And at the same time,

Jonathan Woahn (19:22.786)

get access to it.

Pete Pachal (19:26.621)

as we were just talking about, maybe the training data itself has also been devalued a bit, regardless of where you’re going for it. So it’s kind of like a mix of those things has that caused like less demand on that side. But basically, core question is why have the deals kind of dried up, assuming that’s what you’re hearing too.

Jonathan Woahn (19:43.614)

well, I think the training deals are still happening. I know they’re still happening. I mean, we talked to publishers all the time to talk about the deals that they’re doing right now. They’re smaller. They’re not as big dollar. it seems like they tend to be very, how do I put it? kind of like topically focused. So it’s like, we’re, we’re building a, an agent that is focused on, you know,

Pete Pachal (19:46.954)

Okay.

Pete Pachal (20:05.192)

Okay.

Jonathan Woahn (20:11.724)

medical research and so we’re looking for journals that kind of address this particular domain, right? So it’s not, I feel like we’re definitely seeing a lot less of like the big headline ones that we’ve seen over the last couple of years, but they’re still happening. They’re definitely still happening. It’s just, I think they’re smaller and more targeted and people just aren’t being noisy about it like they were in the early days.

Pete Pachal (20:32.073)

Yeah, maybe it’s just not as much of a news event too. guess there’s sort of a, on the news side of it. Once you land on the moon once, you know, it’s like everything becomes a little road. So I want to tie this back to sort of what we were talking about at the beginning, right? Cause we started out talking about good licensing rails and clean data and having, you know, publisher defined systems for, you know, playing nice with legit operators. And because a lot of these

Jonathan Woahn (20:41.004)

Yeah, exactly.

Pete Pachal (21:02.109)

these gray market companies are presumably doing this, you know, I think it was even in Matthew’s research. He’s sort of like, it can be messy, right? You know, there’s, you gotta get past paywalls and there’s weird stuff when you’re scraping that gets hoovered up as well. And, know, just generally it’s not that reliable or as reliable as, something that is more legit. Does that provide some hope? You know what I mean? Does that, is that like an opening that like, better data cleaner systems?

that’s going to be like, you know, with publisher MCPs and stuff like that. That’s just the demand for that’s going to naturally rise or I don’t know, maybe that’s a little too hopeful.

Jonathan Woahn (21:40.174)

Well, I…

Let me see, I’m trying to think of like how to best answer this question because there’s like some assumptions that we’re kind of building on that I kind of like, I think we need to take a step back and kind of question here, which is like, like one of the questions is like, why are people using, why are these platforms growing? And like, what is it that people are using them for? And like part of it is just because the incumbent alternatives have closed their doors. And so now there’s opportunity, right?

Pete Pachal (21:51.359)

Sure, take them apart.

Jonathan Woahn (22:12.088)

But we’ve, I mean, we’ve started talking with a number of people on the buy side and people who are looking to get access to content legitimately, like they recognize that like candidly, there is some risk that is associated with getting content through some of the gray, you know, the gray platforms. And, and so there are people who are looking to get legitimate access. And part of the challenge that they, that they faced right now is just like, how do you do that? Like, how do you get access to it?

And, um, and I think, I think there’s this, I think there’s this opportunity right now to raise awareness around like the source of where your content’s coming from and is it legitimate and is it creating a sustainable market? And, and so like, just as an example, uh, or an analogy, you know, 20 years ago, going to the grocery store was you just like, I just had to make sure I have food to put on the table.

Like I don’t really care where it comes from. I just need food, right? And what’s been, I think, very positive is in the last 10 years is people have become much more conscious of like, where’s my food coming from? Is it organic? How was it grown? Is this a Georgia peach or is this like a California peach? It’s like, people care a lot more about this. And I think that’s a good thing. And the same thing needs to, sorry.

Pete Pachal (23:33.174)

Yeah, I think it. No, I was just going to say like, there’s a, I don’t know if it’s a sophistication that’s created by supply side, you know, like just having like a lot of supply. I don’t know. I’m not a market genius, the, tell me like, is this an indicator of like an evolving market? And I guess that’s what my question had to do with like, how do we

How is this going to evolve as informational systems get more sophisticated?

Jonathan Woahn (24:01.901)

Yes.

Jonathan Woahn (24:06.828)

And I think that that’s my point here is like, think people need to start thinking about like the term I’ve used is ethically sourced data. Like is this content that I’m working with ethically sourced and is it sustainable? And like, if I’m buying content from a scraper. Like, and it’s not going back to the publisher, how in the world is that publisher even going to be able to continue to create content? Cause if the publisher can’t create content, the scraper has nobody to scrape from. And then you can’t, they have nothing to buy from. And so I think, I think.

Pete Pachal (24:14.069)

All right. Good term.

Jonathan Woahn (24:36.002)

the first step is like, there’s gotta be this awareness on the buy side, particularly around like, where’s my content coming from? And like, how is this creating a sustainable ecosystem? Because I think as that awareness rises, then the question starts to become, okay, well, let’s assume that I actually wanna get my content ethically and I wanna get ethically sourced data, where do I go to do that? And there’s…

not a lot of options at this point in how to go about doing that.

Pete Pachal (25:07.509)

Well, that brings me to like what Cashmere does, right? So can you give me a little bit more on like exactly what your role is in creating? As like, are you someone who cleans up data, create systems, creates a marketplace? Like what is Cashmere’s role in all?

Jonathan Woahn (25:24.268)

Yeah. So our role where we started was ingesting publisher content, getting it ready for AI, structuring in a way that like the AI can now know what the seven habits are instead of 27 habits. And helping to get publisher content cleaned up and ready for use with AI, right? It’s messy. We help clean it up. So that’s the first part. Now, once it’s in place like that, the second piece was how do you manage deployment of that content? How do you get people access to it? How do you help them?

get visibility to it. How do you manage security, authentication, entitlements? Like there’s a lot of kind of moving pieces to like, how do you make that content actually, you know, be consumable? And so what we’ve done is we’ve built all the infrastructure pieces to help connect publisher content with agentic systems. And so our focus has historically just been on like just building that infrastructure and then helping our publishing partners pursue.

the AI opportunities they wanna pursue. What has been interesting though, is as our publisher basis continued to grow, as we’ve started to get more contact, more use cases, more applications, as we started to get a lot more inbound interest from people who are looking to ethically source their content and their data on the buy side. And so we hadn’t done a lot of work on this side because we’ve been very focused on supporting the publishers.

Pete Pachal (26:41.269)

So on the buy side.

Jonathan Woahn (26:50.562)

But now as these opportunities are starting to come in, we are starting to think about, what can we do to help facilitate and expedite getting access for these applications to getting these applications access to ethically sourced content? So we’ve started to do a lot more work on this front. And we’ve got, I don’t know, 15 of these opportunities that we’re running right now. mean, it’s been really fun, to be honest.

Pete Pachal (27:17.205)

And it’s kind of like you’re learning a lot and doing these kind of, I would imagine, content to content buyer sort of handshakes, I guess, from a technical and commercial aspect. And at what point does that become a marketplace or does this become something that is scalable? Like, is that the vision? Maybe it’s a little too early to tell on your side since this isn’t quite the direction you thought you were going. But if that is the case, like,

Like who do you end up competing with in that space? as I know other companies have sort of tried this, they’re all sort of, no one’s done it, done it, if you know what I mean.

Jonathan Woahn (27:56.802)

Yeah, it’s, mean, Pete starting, starting a marketplace is really, really hard. and you’ve got to like, I mean, I think there’s, there’s a marketing component to it. There’s a technical component to it. There’s a lot of luck. I think that sits like being in the right place at the right time with the right people. And so like, I think there’s a version or a future where like what we’re doing with cashmere, I think it could become a marketplace. but at the moment, what it feels more like we’re doing at

Pete Pachal (28:02.74)

Mm-hmm.

Jonathan Woahn (28:25.878)

like this very instant as it feels more like we’re like brokering relationships, right? Like we’re, we’re initially just taking like, we’ve got this super powerful platform, this super powerful technology that sits under the hood of what we’ve built. And now we’re just trying to help, like we’re trying to help publishers monetize their content. Like that is like our goal, right? and if we can help do that by saying, Hey, we’ve got people who want your content. And what’s great is now we’ve got, you know, we have someone that comes to us and says,

As an example, we were talking to a South Korean hardware manufacturer who, not South Korean, South Asia, a hardware manufacturer who has like a Siri competitor. And they’re looking for access at inference time to news content. And they’re looking for kind of like, you know, some kind of lifestyle type content. And so we’ve got a handful of publishers who we’re working with who do this. And so now what we can say is like, well, publisher, we were working on this other thing with you to start with.

But now we’ve had someone coming and asking us for this, is this something you would be interested in getting involved in? And so now like we’re able to start taking this network of publishers that we have and start bringing them together to, to, bring that content into these applications side. So it does feel a lot more like brokering arrangements right now than like a marketplace, but you know, maybe at some point it could turn into that, but that’s not like what we set out to build at the moment.

Pete Pachal (29:51.446)

All right. Yeah, the future is long. Let me ask you like, I guess, probably a pretty important question, which is like, what does fair pricing look like in an AI content marketplace? What are the factors? Obviously it’s gonna be different depending on the content and the people involved in the brokering, but like, again, what are the dimensions that would govern pricing?

Jonathan Woahn (29:54.734)

The future is long. Yeah.

Jonathan Woahn (30:16.012)

Yeah, this is a fantastic question. I’m actually working on some writing up some content around this right now, because I’ve been doing a lot of research on this.

I think there’s fundamentally two factors to think about. The first is the intrinsic value of the content itself. And then the second is the particular use case of how that content is being deployed. And so like, if you kind of break each of those apart, like at a high level, on the content side, the intrinsic value, you know, there are different categories of content that we can look at. So,

Content that is from user generated content, like something you might find on Reddit doesn’t have the same intrinsic value as like a market research report that one of the sell side brokerage firms has put two years of research into, right? Like the value on a per token basis is not like dollars to dollars. Did you have a question? Sorry, I was just looking at.

Pete Pachal (31:17.427)

No, no, no. I’m just following along. I’m right with you. It’s not dollars to dollars.

Jonathan Woahn (31:21.194)

Okay. So there’s these different kinds of categories of content, right? So you’ve got like, there’s like open web content. You might have news, you might have a lifestyle. might have books. You might have market research, market intelligence, scholarly research, right? Like there’s different vertical kind of categories of content that each have. There’s value depend on in each of those. And then even within that, there’s difference in how you think about like front shelf content.

versus back shelf content, right? The stuff that is driving all the attention to your website. And then once you get them in, then you’ve got this other stuff that you might be able to perform to keep them there. So there are some metrics and like, I can get pretty deep on this if it’s helpful, but like.

Pete Pachal (32:06.815)

So far, I’ll let you know when we want to get back to the surface. But go, this is great.

Jonathan Woahn (32:10.486)

Okay. So there’s this intrinsic value of the content itself. That is one of the factors in how to think about pricing. So then if we kind of look at the second factor, which is like the use case, there’s also use cases that dictate the value of that particular content. And so when I say use case, it’s like, how is the AI wanting, what are they wanting to do with that content? And the analogy,

Pete Pachal (32:35.764)

Right.

Jonathan Woahn (32:39.054)

kind of that I’ve drawn for from this has come from the music industry. And so like if you go to ASCAP, you can go to their website and literally like right on the homepage, it’s like, what is your use case? Like, what are you using the music for? And within that use case, you can say, I wanna, I have a restaurant and I want to, you know, have music in my restaurant. So then you click on this and it shows you, it’s like, well, are you a karaoke bar?

Are you wanting to play it in the elevator? Are you wanting to play it as background music? Are you a music like, like what is your actual use case? And then depending on that use case, the way that they structure and price that license looks very different, right? Like it’s, it’s going to cost more to if the music is much more of a focus of what you’re doing versus if it’s just like background music. And so on an AI, we can think of it very similarly to say, well, what is the use case that you’re wanting to use this content for? And.

If you’re just doing like a simple, like chat, like rag chat application, like the value of that content, like it’s probably pretty substitutable. Like, you know, if it’s just like a generic search engine, but if you’re doing like a verticalized search engine where now you’re focusing on, like as an example, like, open evidence, like, I’m not sure if you’re familiar with this, but like open evidence is like pulling in all of this academic research.

Now what you’re doing is you’re providing a use case that has a very specific audience with a very specific need and is very specific kind of outcome they’re looking to drive. And so, you know, what you charge for a chat GPT at $20 a month and like what you get access to there is very different than what you charge, you know, $200 a month for like a deep research agent. That’s like very vertically specialized, right? And so you got to kind of look at these two things to say, what is the intrinsic value of the content and what is the particular use case?

Pete Pachal (34:23.165)

Hmm.

Jonathan Woahn (34:30.7)

And that gives you an idea of what the value is for that content, for that particular use case, and how you think about pricing.

Pete Pachal (34:38.239)

Got it. So a lot of it matters just in terms of like the outputs. It feels like the outputs is always kind of like a big determining factor in all this.

Jonathan Woahn (34:46.264)

Correct.

Pete Pachal (34:48.767)

Cool, okay, so we talked a little bit once just over email about the agent layer for publishers, which I think is an interesting way to sort of think about all this. And what basically, I’m curious what you think about what good product strategy, I guess, looks like for a publisher as they think about that agent layer.

And I know we’ve sort of like touched on a lot of this in terms of just clean data and et cetera, but how do you make sure it’s something well-designed that’s gonna both serve you, the people you broker with, but also the ultimate users, right? Whether they’re readers or analysts or what have you.

Jonathan Woahn (35:35.724)

Yeah. So is the question here just like, are, as you think about designing, like if I’m a publisher and how I’m designing, like what am I doing from a product perspective?

Pete Pachal (35:46.666)

Yeah, because it’s like, feel like when we talk about AI, the most people default, it’s a chat bot, Like you don’t like have a chat bot in front of a public site. And honestly, it’s sort of like, that’s kind of what a lot of experiences are now from a user perspective. like, obviously like having an MCP and that data available just from a prompt that you can, whether you deliberately ask for the data from this publisher or it’s just implied.

Jonathan Woahn (35:51.374)

Mm-hmm.

Pete Pachal (36:16.629)

sort of gets it. You know, I just feel like I’m not sure maybe if this is not, I don’t know if this is more incumbent on the publisher or the person using it, but I guess if you are the publisher, like how are you thinking about that? And I guess it can vary widely depending on what the person is doing with it, but how do you make sure that it’s all those things are available?

Jonathan Woahn (36:36.46)

Yeah, it’s,

Pete Pachal (36:40.373)

Like what are the top three things you would recommend someone to get started with before you even get in there?

Jonathan Woahn (36:49.166)

Yeah. Well, think, I think the first, the first question is like, is, is kind of, I mean, it’s, kind of like, what is your AI strategy here? And do you want to host your own infrastructure and do you want to host your own content? And then the question that I have that falls behind that is like, why, like, like, what is it you’re wanting to do with it? Um, and, and so then, you know, like, but if we go back to that first question of like, you know, what is your AI kind of strategy? Um,

Pete Pachal (36:58.099)

Right.

Pete Pachal (37:09.235)

Yeah.

Jonathan Woahn (37:19.224)

Your strategy might just be, we’re going to make our con, we’re going to license our content on a consumption basis and we’re going to, you know, deploy it through our website and use some of the kind of gateway platform infrastructure that currently exists. Right. And so in that case, if that’s your strategy, then a lot of that comes down to making sure that you’re, you are, structuring your content in such a way that makes it easier for the agents to consume it and understand, you know, how to, how, how your content is actually structured.

So you hear lot about, you know, people talk about having a, you know, there’s like the HTML version that people see and then you can have like a markdown version that people don’t see this is on the scenes that like, you know, that an agent might get access to through website.

Pete Pachal (38:00.822)

Right.

Pete Pachal (38:06.803)

Yeah, totally. No, no, no, I think it is. But I think it’s sort of like thinking of, as a publisher thinks about playing in this world of AI, you know, a lot of it has to do with defense. And that’s kind of sort of trying to get not quite offense, but like, I want to have a presence in this legit, you know, quasi marketplace or forming marketplace.

Jonathan Woahn (38:10.264)

I don’t know if I’m answering your question, so I’m like.

Pete Pachal (38:33.523)

What does that look like in terms of both approach and execution? That’s kind of what I was thinking about. But I also, know, as, know, like I haven’t seen too many pages for maybe these exist and I just haven’t seen them, but like, like a page on a publisher, it’s like, Hey, if you’re interested in like, our A our A our agent and our MCP, you know, call us or like, you know, put them in this application or et cetera.

Jonathan Woahn (38:55.63)

Yeah, here’s how you get legit access to this.

Pete Pachal (38:57.769)

You know, yeah, like kind of just putting it out there. Maybe there should be more of that. I don’t know. I haven’t seen that. Maybe that exists. You tell me.

Jonathan Woahn (39:09.356)

Well, I think it does exist. I mean, this is like what Tolbit is doing. Like anybody who’s using Tolbit, like you get an MCP server by default by hooking up with them, right? I think what happens on the other side of this, that’s a little bit challenging though, is just around the discovery layer. And if you are a legit buyer who’s wanting to access this content. So just as an example, there’s a consulting group that we’re talking to who wants again, legitimate access to high quality content.

Pete Pachal (39:11.881)

Yeah. True.

Jonathan Woahn (39:39.306)

They don’t want to buy it through the scrapers, but they have hundreds of sources of content that they’re wanting to get. And so you can imagine that, you know, if they want to get it from all of these different publishers and their content, like what are they going to do? Go around to every single one of them individually and sign up for it and integrate it. Cause now you’ve got hundreds of data processing agreements you’ve got to go through. You’ve got to make sure they’re all like have information security covered. You’ve got to make sure that

They’ve, you figure out like, what is the pricing going to look like? And like, to your point, I don’t see a lot of websites, a lot of these publishers that are saying, Hey, you know, we might have our MCP, but like, here’s how our content is. Here’s how you get access to it. Right. Like there starts to become this, kind of fan out, you know, power scale problem here where.

If you want legitimate access, how do you actually go about doing this? And you’re right, today as it is, there’s not an easy way to go about solving this at this time.

Pete Pachal (40:41.939)

Yeah, it just seems like it’s definitely a manual kind of roadblock, or at least a manual process right now that tends to be a roadblock in many cases anyway. This has been great. Just to wrap up, I always try to end on the same question, which is that you’re looking, projecting forward. There’s got to be things you’re worried about and hopeful about. Pick one from each of those columns and give it to me in any order.

Jonathan Woahn (41:09.198)

I’ll start with what I’m worried about. I think the thing that I’m worried about is just, I worry that the kind of incumbent experience is going to be severely negatively impactful for publishers. And what I mean by this is like, you know, I like, there’s a lot of analogies and parallels we can pull from the music industry, but

You know, when Steve jobs came out and said, we’re going to sell a single track for 99 cents and here’s how the economics are going to work. And here’s, know, how we’re going to deploy it. Like it reshaped the industry to kind of think in a different way. And it really, I think it expedited the ability to kind of adopt a new business model. And what I worry about with AI and with a lot of publishers is just, you know, there’s a lot of incumbent ways that we’ve done business historically. And there’s not like a really clean unified way that says, Hey, this is how it works with AI.

And I think what ends up happening in that instance is I think the buyers end up dictating a lot of the terms and then publishers end up becoming price takers. And that worries me because I think we have an opportunity here as publishers to effectively reset the table and say, this is how we want it to work and this is how it’s going to be sustainable for us and for you going forward into the future. So that’s what I worry about. What I am hopeful about is…

We continue to see that AI needs access to fantastic content. And there’s nobody better at creating that fantastic content than publishers that they’ve proven over decades and centuries. And there’s a lot of publishers who are really keen and really looking forward on trying to figure out how to do this. And so it’s very encouraging to be working with some of those to see the strides that they’re taking and how forward looking they are on this. And it gives me hope. It gives me hope that,

If we can help to show that these use cases work and help to provide some case studies that like this is sustainable and this is a big opportunity, my hope is that it will address my worry and it will bring the rest of the publishers along.

Pete Pachal (43:15.219)

Nice. It’s quite a vision. We’ll leave it there. Jonathan, thanks so much for stopping by and sharing your thoughts.

Jonathan Woahn (43:21.752)

Thanks Pete for having me, appreciate it.

The post The Scraper Economy is already here. Publishers just aren’t getting paid. appeared first on The Media Copilot.

How AI search is rewriting the rules of discovery

Pete Pachal — Tue, 26 May 2026 12:00:00 +0000

The AI search traffic apocalypse is a story most working journalists, marketers, and comms pros know by heart, and I’ve written about it plenty. The premise is simple: If your business runs on funneling readers from search engines to a website, AI summaries are cutting that funnel off at the top, sending the audience instead to a paragraph the bot wrote about your work rather than the work itself.

The compensation question hangs over all of it. Publishers may eventually get paid for the material that AI systems are scraping and citing. They may not. But the more pressing shift is happening in plain sight. The battle for attention is moving toward new victory conditions, namely whose information gets cited most prominently in an AI summary. AI presence doesn’t replace traffic, but it does function as the new proxy for relevance and authority.

In my first Fast Company column, I argued the incentive system this creates is healthier than the one we’ve lived under for over a decade. Search and social trained media to chase engagement, which produced the fire hose of listicles, outrage bait, and formulaic explainers like “What time is the Super Bowl?” If AI systems are the arena, and if they really do reward well-sourced, domain-specific content more than social heat, that could lead to a resurgence of good journalism, at least directionally.

Enough time has passed to start checking the theory against actual data. AI search engines have been operational for more than a year, usage is rising fast, and researchers are starting to publish on what these systems actually cite. The early signal is encouraging, but the caveats are real.

AI’s reading list

A pair of recent studies looked at millions of AI citations, the term for a source that an AI summary both names and links. The data found that AI systems treat LinkedIn as one of the most authoritative sources on the internet. Research from Meltwater, a communications intelligence company, showed LinkedIn as the second most-cited source overall in AI summaries (after YouTube), and a separate study from Semrush, a search-data analytics company, concurs, also putting it at No. 2, closely behind Reddit.

The Meltwater data also point to why LinkedIn is a decent indicator of substance. Individual members (rather than brand or company accounts) drove most of the citations, structured content like newsletters and posts performed best, and more than half of the citations went to members with fewer than 10,000 followers. Likewise, Semrush found that the most-cited LinkedIn posts had only modest engagement on the network itself. That’s strong evidence against a simple popularity model.

The harder finding is what AI systems do when structure is missing. When you drill down into academic papers that zero in on exactly how large language models prioritize information, like this one from the Canadian AI company Cohere, they show that LLMs will miss key facts when an article lacks clear titles and headings. A separate paper from Stanford University goes further, showing AI search systems strongly favor the beginning and the end of documents over the middle. If the meat of your reporting sits in paragraph seven, the bot may never reach it.

Put those findings together and AI systems look as gameable as search and social were, just along different axes. An article optimized for machines, with declarative ledes, clean Q&A, and consistent titling throughout—but otherwise empty of substance—could theoretically outscore a piece with original reporting that lives in the middle. AI systems reward retrievable substance, not necessarily the most insightful or information-dense content.

Visibility to AI engines isn’t enough on its own. You have to lead the bots to the good information instead of hoping they’ll find it. This is the whole idea behind GEO, or generative engine optimization, and it can feel at odds with what makes good writing good. Clever titles, narrative hooks, the slow backing-into of a topic through a scene: humans love that work; machines mostly don’t.

The human edge in machine search

Look back at which sources are ranking highest: LinkedIn, YouTube, Reddit. That mix suggests the best content is a blend of machine-friendly formatting and the human element. AI doesn’t always cite the most engaged posts, but the Semrush data also shows that frequent posting and an established following still help. LinkedIn’s own internal guidance points in the same direction. So engagement still matters, just less directly than it did in the previous era.

Demoting raw engagement is progress. The structural bias is something working journalists and content creators can put to use. It signals that content based on original reporting or insights needs to do several things: explain concepts clearly and quickly, include machine-friendly structures like subheadings, and connect the dots with other sources the AI is reading by referencing them by name.

The play, then, is to make good work easier for machines to read without sanding off what made it valuable to humans in the first place. The next incentive system will have its own bad habits, and there’s no doubt many people will try to exploit them. But if AI search continues to weight original facts, named sources, clear context, and demonstrated expertise over outrage and raw engagement, that’s an opening worth taking seriously. The winners in an AI-mediated future shouldn’t simply be the loudest accounts or the best-formatted posts. They should be the people who know something real and can demonstrate its worth to both audiences they’re now writing for, human and machine.

A version of this column appears in Fast Company.

The post How AI search is rewriting the rules of discovery appeared first on The Media Copilot.

YouTube is now the No. 2 most-cited social platform in AI answers

Lornah Ngugi for WebFX — Wed, 20 May 2026 13:07:04 +0000

AI search engines cite YouTube videos because the platform often provides structured, in-depth information that AI systems can extract and reference in generated answers. Long-form videos, transcripts, timestamps, chapter markers, and detailed metadata make YouTube content especially easy for AI systems to analyze.

WebFX observed a recent study that found that the platform accounts for 38.1% of all social media citations in AI-generated answers. This makes it the second most-cited social platform across major AI search engines, including Google AI Overviews, Google AI Mode, Perplexity, and ChatGPT.

This shift matters for marketers and content teams because AI-generated answers increasingly shape how users discover information online. As YouTube citations grow, well-structured video content can directly influence brand visibility in AI search results.

Why do AI search engines cite YouTube content more often?

Courtesy of WebFX

AI search engines cite YouTube because long-form videos contain detailed explanations that can be converted into text through transcripts. These transcripts give AI systems structured information they can extract, analyze, and reference when generating answers.

YouTube also hosts a massive library of educational and instructional content covering millions of topics. As a result, AI platforms often treat YouTube videos as knowledge sources, not just entertainment content.

This is evident in the fact that many of the videos cited in AI answers come from content most viewers have never encountered. According to the analysis, 40.83% of AI-cited YouTube videos had fewer than 1,000 views at the time of the study, while 36% had fewer than 15 likes.

Courtesy of WebFX

However, YouTube video AI citations vary across platforms, with Perplexity and Google AI Overviews accounting for roughly three-quarters of all observed YouTube citations in AI-generated answers.

Here’s a breakdown of the share of total YouTube citations across different AI platforms:

Perplexity: 38.7%
Google AI Overviews: 36.6%
ChatGPT: 4.4%
Gemini: 0.2%
Microsoft Copilot: 0.5%

WebFX

What kind of YouTube videos do AI search engines cite?

According to the study, the most frequently cited YouTube videos by AI search engines include:

WebFX

Let’s unpack each type of YouTube video below.

1. Long-form educational videos

AI search engines overwhelmingly cite long-form, reference-style YouTube videos that explain topics in depth, providing AI systems with enough context to summarize.

The dataset reveals that 94% of YouTube citations in AI answers come from long-form videos, not short-form content.

That trend contrasts with how many brands currently approach video marketing. In the past few years, marketers have prioritized short-form formats such as YouTube Shorts, TikTok videos, and Instagram Reels to maximize reach, engagement, and algorithmic distribution across social platforms.

But AI citations are changing that because they’re continually citing long-form videos that behave more like mini knowledge resources, for example:

Tutorials
Product explainers
Detailed walkthroughs
Documentaries
Vlogs
Interviews
Lectures

2. Videos with time stamps and chapter markers

Video structure also affects how frequently a YouTube video appears in AI-generated answers. Videos that include time stamps or chapter markers allow AI systems to reference specific segments rather than the entire video.

Courtesy of WebFX

When Google AI Overviews or Google AI Mode cite time stamped videos, they often link directly to individual sections. This structure effectively turns a single video into multiple citation points, expanding the number of opportunities for AI systems to reference it across different queries.

3. Newer, trend-relevant videos

Another factor that appears to influence AI citation patterns is how recently a video was published. The study found a weak positive relationship between recency and citation frequency, indicating that newer videos were cited slightly more often during the observation window.

This pattern is most noticeable in queries where fresh information matters, such as searches for “latest,” “new,” or a specific year, like “2026 fashion trends” or “top Amazon products for 2026.” In these cases, AI systems often favor more recent sources when generating answers.

4. Videos with clear metadata and structured descriptions

The analysis found that videos with more detailed descriptions were cited slightly more often than those with minimal descriptions. This suggests that clear summaries and structured metadata help AI systems better interpret a video’s topic.

Citable YouTube video descriptions should:

Explain what the video covers
Highlights key concepts,
Include structured elements such as chapter lists, keywords, or relevant terms
Include hashtags for additional topical signals about the subject of the video

Courtesy of WebFX

What YouTube content AI systems rarely cite

The analysis found that several common YouTube video optimization features show little measurable influence on whether a video gets referenced in AI-generated answers. Some of these include:

Video popularity signals: Metrics such as views and likes have little effect on how often a video is cited by AI platforms.
Channel size and subscriber counts: Larger audiences did not consistently translate into higher citation frequency.
Total number of channel videos: While a larger library increases the number of possible citation candidates, it does not directly increase the likelihood that any single video is cited.
Video duration alone: Simply making longer videos does not guarantee citations. The structure, relevance, and clarity of the explanations matter more than length by itself.
Title length optimization: The dataset found no meaningful relationship between title or description length and citation frequency.

WebFX

How to optimize YouTube content for AI extraction

If AI search engines increasingly treat YouTube videos as reference sources, content teams may need to rethink how they structure video content. The patterns identified in the study suggest that videos most likely to appear in AI-generated answers share several characteristics:

1. Focus on long-form explainer content

AI systems most frequently cite videos that fully explain a topic rather than briefly introduce it. For many topics, these are videos in the five- to 20-minute range that can be broken down into digestible chunks.

Long-form videos also tend to produce clearer transcripts because they include structured narration and complete explanations. This makes it easier for AI systems to interpret the content and identify specific segments that answer a user’s query.

Effective transcripts for YouTube AI citations include:

Clear spoken explanations, not just visuals or background narration
Structured sections or chapters that organize the topic logically
Natural use of keywords within the narration
Complete explanations of a question or process

2. Structure videos with chapters and time stamps

Videos that include time stamps or chapter markers are more likely to be referenced and cited by AI search engines. AI systems interpret time stamps, especially ones labeled in user-friendly language as subheadings, making your videos more extractable.

In fact, 78% of time stamped videos show a higher likelihood of being cited again. Additionally, structured video content also allows for more YouTube AI citation opportunities across different questions, particularly within Google’s AI search surfaces.

3. Treat descriptions as structured metadata

Video descriptions often serve as metadata that help AI systems understand what a video covers. Descriptions that clearly summarize the topic, list key concepts, and include relevant terms make it easier for AI models to understand a video’s content.

Chapter lists, keywords, and supporting links can further clarify the subject matter for AI systems.

4. Keep content current when topics evolve

Recency can also affect YouTube AI citation visibility, particularly for queries where users expect up-to-date information. For industries that change quickly, such as AI tools, software updates, marketing tactics, or product comparisons, regularly updating or publishing new videos can help maintain relevance within AI search ecosystems.

This story was produced by WebFX and reviewed and distributed by Stacker.

The post YouTube is now the No. 2 most-cited social platform in AI answers appeared first on The Media Copilot.