Bot traffic passed human traffic on the web this year, and most publishers are still trying to keep the bots out. TIME went the other way. On July 28 it started serving ads to them: brand-verified facts, labeled as sponsored, placed on the machine-readable versions of its pages for the roughly 70 AI agents it lets in. Ally Bank and Project Management Institute bought first. Perplexity, a TIME licensing partner, called the units deceptive.
Listen or watch:
Mark Howard, TIME’s chief operating officer, is back on The Media Copilot to explain what a month of agent ads has taught him. The infrastructure came first: TollBit and ScalePost on the CDN, an allow list of agents that get through, a paywall for everyone else, and a markdown page built for machines. Then the product, developed with the ad tech startup Mobian, which measures what a brand says about itself in paid media, runs dozens of prompts through the major models, and finds where the AI is invisible, unfavorable, or simply wrong. The agent ad exists to close that gap.
“We would rather be at the table and learning as we go, as opposed to just simply waiting for others to pave the way.”
The most surprising claim in the conversation is about memory. TIME sees more AI referrals to some of its articles than it sees AI citations, which should be impossible. Howard’s explanation is that the models are storing what they scrape and retrieving it later, and his conclusion follows from that: an impression served to an AI agent may be worth more than one served to a person, because the agent keeps it.
“You can’t have more referrals than you have citations. So clearly they’re storing the information.”
Pete pushes on the hard parts. What does an advertiser actually buy when nothing is guaranteed? If a model can cleanly separate the ad from the editorial, why wouldn’t it just throw the ad away? Who counts as a legitimate crawler? And does Perplexity have a point? On that last one, Howard says TIME met with Perplexity, walked them through the markdown page and the HTML source side by side, and hasn’t changed its position. Perplexity’s crawlers, he adds, haven’t slowed down.
“Not only is this not deceptive, but it is the most transparent and cleanest way to present to an AI crawler the information.”
This episode is part one of a two-part look at ads for bots. Part two, next week: Oasy co-founder Choy Travers on building the same idea as a product for every publisher.
In this episode
Pete and Mark discuss:
- Why TIME chose to learn by doing instead of waiting for the market to settle
- The infrastructure underneath agent ads: TollBit, ScalePost, and a 70-agent allow list
- The markdown page built for machines, and what an agent ad actually is
- Which bots show up, and what they’re there for
- The TIME100 Creators moment that made the case for the product
- Mobian’s method: visibility, favorability, and accuracy across the major models
- Why TIME believes the agents remember what they scrape
- One month of data: causal lift and model impact over time
- What the advertiser is buying when there’s no guaranteed placement
- Whether the models can read the sponsored label, and what they do with it
- The Perplexity dispute, and what OpenAI thinks
- What counts as a legitimate crawler
- Pay-per-crawl as both the biggest worry and the biggest hope
About the guest
Mark Howard | Chief Operating Officer, TIME | LinkedIn | time.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.
Produced by The Media Copilot.
Music: “Favorite” by Alexander Nakarada, licensed under CC BY 4.0
All rights reserved. © AnyWho Media 2026
TRANSCRIPT
[0:02] Pete Pachal
Hi, welcome to the Media Copilot. It’s a podcast about how AI is changing media, news, and communication. My name’s Pete Pachal. I covered tech for a long time as a journalist, and now I have deep conversations with the media people, the builders, and the creators who are answering the question, how will we get information in the future? And how will that change the jobs and industries whose business is information, especially media? One of the biggest questions in media right now is what happens when the audience isn’t necessarily human anymore? Publishers have spent decades building businesses centered around people coming to their websites. Search got them traffic, people read stories, saw ads, clicked some affiliate links, subscribed, and hopefully came back. AI is changing that bargain. Chatbots can crawl a publisher site, absorb the reporting, and answer a user question without sending much of anything back.
And increasingly, AI agents may be doing research and choosing sources and narrowing options before a human ever sees them. So if bots are becoming a meaningful audience for publishers, can publishers actually make money from that audience? My guest this week is trying to find out. Mark Howard is the chief operating officer of TIME. When Mark was last on the show, we talked about TIME’s push to become an AI-forward publisher. That included licensing and distribution partnerships and the TIME AI agent, which gives readers a new way to interact with more than a century of TIME journalism. Well, now TIME is testing a different kind of business model. Through a partnership with Mobian, an ad tech startup, TIME has begun putting sponsored content into machine-readable versions of its pages. The ads are labeled and designed for AI crawlers rather than human readers.
Ally Bank and Project Management Institute were the first announced buyers. The theory is that an AI system may absorb the sponsored material and use it when somebody asks a relevant question. That would turn a bot from an uncompensated scraper into something closer to an audience you can actually advertise to. I’m really fascinated by this idea because it attempts to solve one of the biggest problems publishers have with AI right now: lots of machines consuming their work and very little economic value flowing back.
But this gets complicated very quickly. Among the questions: what exactly is an advertiser buying if there’s no guaranteed appearance in an answer? How do you measure whether the message actually influenced an answer? How should advertising be disclosed? And what’s the risk of the label never reaching the person reading the answer? And what happens if companies building the AI systems decide they don’t like this, they don’t want publishers doing it at all? That last question became
especially important when Perplexity called TIME’s approach deceptive and said it had blocked the sponsored material from influencing its index. TIME and Perplexity have been partners, which makes this more than a theoretical disagreement. So this is a conversation about whether ads for bots are the beginning of a real market, what the early data actually says, and whether a publisher can monetize its authority without weakening the trust that makes the product valuable. It’s gonna be a good one. Before that, though, 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. Mark Howard, welcome back to the Media Copilot.
[3:33] Mark Howard
Thanks, Pete. It’s wonderful to be back.
[3:36] Pete Pachal
Nice. So I want to get into all the thorny questions, thorny or otherwise, that I just mentioned, but first I want to see where things are at with you and TIME. Last time you were here, I think you described TIME’s AI strategy as a learn-by-doing approach. So what have you learned, and how did it lead you to this idea of ads for bots being essentially the next experiment worth running?
[4:05] Mark Howard
That’s exactly right. We’ve adopted this mindset of we would rather be at the table and learning as we go, as opposed to just simply waiting for others to pave the way. And I think that that has led us to this moment in time right now that you so elegantly described in the setup here, about where we see an opportunity. Everybody in the media industry has been observing for the last several years that fundamental shifts are occurring in terms of consumer behavior, the role of AI answer platforms, the emergence of agents across the web, and the explosion of those agents now being more dominant, or a higher percentage of the total internet than human-based traffic. And so there’s a number of variables that have all
emerged to get us to this moment in time. What’s really exciting and most interesting is we only launched the first agent ad, an ad targeted specifically to an agent who came to Time.com, one month ago. July 28th was the launch date. But since then, quite a bit has happened in terms of how we’ve evolved our thinking about this product offering. You can only imagine the dozens of conversations that we’ve had with various brands and agencies that are super interested in this. We’ve also had a lot of conversations with other media companies who are curious about how exactly we’re doing this and how the partnership with Mobian has created this opportunity for where we are. So there’s a lot to get into.
But at the core is exactly what you just said. We started on this journey back in 2023, focused on infrastructure. And at that point, in partnership with our CDN, we layered on TollBit and ScalePost to start tracking what agents and bots were coming to the site, what they were coming for, and what content they were consuming. That foundation was critical and is critical for how we’ve been able to get to today. That also led to the ability for us to do a number of licensing deals. Those licensing deals have created strategic partnerships for us with companies like OpenAI, Amazon, and Perplexity, as you mentioned. So we’ve got that foundation in place, and it’s enabled us to learn quite a bit through those relationships as we’ve gained access to their partnership teams as well as their senior executives through a series of meetings over the last several years.
So that has been critically important to framing some of the thinking that’s enabled us to get to this point. Our last conversation focused a lot on the release of the TIME AI agent. That was a huge process for us to take a 103-year archive, create a vectorized database, be able to create a tool that would enable people to interact in chat format with our tremendous archive of information chronicling the history of the world over that century-plus.
And then with that, we’ve now found ourselves in this position where, in spirit with Media Independence Day that Cloudflare operated on July 1st, 2025, we embraced that blocking initiative. We further evolved that to create now what is an allow list. We have about 70 allowed or approved agents and bots that we allow through into the site, and we route all of the unapproved bots and agents to a paywall, which they have the opportunity to pay a small fee in order to come through. The vast majority, of course, do not, which is probably a topic we can get into later in terms of pay-per-crawl. But with that, we also several months ago introduced the markdown page, or the machine-readable page, which is now where, when an agent or bot arrives at time.com
that is on the approved list, they are routed to this machine-readable page versus the HTML page that you and I and everybody else who would visit time.com would experience, that has the HTML, the JavaScript, the photos, the videos, all the things that we all believe makes for compelling storytelling and better consumption of content and information. The agents, of course, do not care about any of that. They want the metadata and they want content.
And so we started separating that out several months ago, and with that separation, introduced this concept of a new surface environment that is targeting this new, hugely valuable audience, which are these agents and bots that are coming in increasingly high numbers and growing every month. So we view that as a big opportunity, and we view the credibility and the authority of TIME journalism, which is why they’re coming in the first place, as a positive for being able to even launch this product, which we call agent ads, which at their core are brand-verified facts presented in a very machine-readable format, so that it is easy for the agents to understand exactly
where the ad begins and ends, what is the ad, and how that is separated from the editorial product. And we now have a number of proof points that can verify that each of the major LLMs can decipher what is the ad and what is the content.
[10:01] Pete Pachal
Nice. I want to get to that in a second, but I just want to reflect back a bit, but also fully understand. So what are the boxes to check, the things you need in place? You went over some of it. Number one, it sounds like it’s simply a perspective: don’t think of bots necessarily as just distribution infrastructure, but also as an audience, as advertising inventory, essentially. Number one, understand that. And then once you have that, moving some technical pieces with the things you mentioned about having the machine-readable version of your articles, or essentially making sure you’re paying attention to what the bots actually read per page, and then
what we’re gonna touch on very quickly is measurement, right? There’s gotta be some way to verify, measure, and present to the advertiser what exactly you’re buying. So first of all, what am I leaving off that list? What are the boxes to tick for anyone that wants to go into this that you need to put in place before you even attempt this? And then let’s get into: what is the advertiser actually buying, and what can you verify through the end-to-end AI scraping and presenting process? How is that trackable? Really want to get into that, but please.
[11:42] Mark Howard
It starts with the daily, weekly, monthly observation of which agents and bots are coming. Are they coming for AI assist or web assist, meaning real-time retrieval? Are they coming for indexing? Are they coming for training? Which companies are they coming from? They are supposed to, at the CDN level, declare who they are and what they’re there for. We believe that the folks on the approved list are accurately declaring who they are and what they’re there for. We’ve observed interesting patterns that a vast majority of them will come and they’ll go to our homepage or to various hub or index pages where all of our content is presented to them in an easy format for them to be able to then go on to the articles and consume the content. So we’re studying natural behaviors of the different
agents that are coming. There’s also differences in terms of whether Anthropic is coming for real time; we actually see that they come more for training. We see that ChatGPT is coming more often with AI assists, meaning more real time. We look over time to see: do the patterns differ between the types of content that the different AI agents are coming for? We see spikes in the traffic a couple months ago when we launched the TIME100 Creators list, which was really one of the aha moments for us. We saw that there was this huge spike in AI assist agents coming, so real-time agents coming, operating on behalf of people, obviously, that were coming to the hub page of the creators launch, and then going to the individual creators that were on that list. And so
that sparked for us this idea. Ally was the sponsor, is the sponsor, of the Creators, in terms of when you and I go to the site, you’ll see Ally ads on that page. They wanted that deep association with the types of people that are on the page and ideally the kinds of people that would be interested in understanding and reading about the people that we named to the TIME100 Creators list. Same time, we saw this huge spike in these AI assist agents coming to the markdown version of that page. And that was really the big aha moment for us: that there was a big opportunity here for us to use that markdown page as a new piece of inventory that is highly valuable, because those agents were operating on behalf of people that were coming in huge numbers over the first several days that the list launched to get that information.
And so that’s when it really crystallized for us that we should go to Ally and talk to them about, you should also have your version of a machine-readable ad on that markdown page, the same way that we’re presenting the content to the agents on behalf of TIME editorial. So with that, we quickly sprung into action. We had been having these conversations, as you mentioned, Project Management Institute and Ally, the two that we launched with. We’ve been talking with Project Management Institute for quite a while, as well as a number of others, about this premise, but it was really that moment where we saw that behavior of the agents that made it really clear that there was this immediate big opportunity. And both of those partners were tremendous in terms of moving quickly with us to be able to capitalize on being first to get their ads live. And we are now serving, in
the cases of both, they’ve received over a couple million agent impressions that we’ve been able to serve on these markdown pages on their behalf. So one, it was understanding the agent traffic. Two, it was having that infrastructure in place where we already had the markdown pages ready to go and we were observing the behavior of the different agents on those pages. Three, it was partners who are dealing in, every brand that we’ve spoken with, they are working through what AI visibility means, what GEO and AEO is, how to measure it, how do they show up, what are their opportunities. We know a lot of the different vendors that are out there, and we’ve met with them, that are selling them opportunities to do that form of measurement. But we actually are taking a little bit more of a nuanced approach to what
we think the bigger opportunity is. And this is again where our deep partnership with Mobian, who we’ve been working with for three years now, first on the brand safety side and then on the measurement and AI persona targeting, and so deeply integrated into our ad sales. What Mobian does is Mobian measures all the different paid advertising that exists on the open web and across social platforms. And they build these very sophisticated brand briefs on each advertiser about what they are in market saying about themselves. And so how they are using their paid media to convey to people who they are, what their products are, what their product offerings are, their brand values, the benefits of working with them. That is the baseline.
And so we start with that premise of, what is their advertising saying? Then we use that to generate dozens and dozens of prompts that we feed into each of the different LLMs to then generate: well, what does the AI say about you? Is the AI saying the same thing about you that you are saying in your advertising? And where is there a gap? And that gap is where we think the opportunity is for the agent ad offering that we brought to the market. Inevitably there are gaps. The AI is not conveying, whether it’s talking about them from a visibility standpoint in general, whether they’re showing up; when they are showing up, whether the AI is speaking favorably about them; and when it’s showing up, are the brand facts accurate. So that framework of
visibility, favorability, and accuracy is really how we frame the conversation with each brand. And that represents how we then tailor what the agent ads are going to address.
[18:36] Pete Pachal
Yeah, and that’s interesting, because you have that gap, and the bigger the gap, the bigger their opportunity. And then you have a good case for them to do this. But it also sounds like the data that you have can inform who should do this, and also what content specifically those ads might be most performant on, right? And so that leads to my next question, which is: how important is editorial-ad alignment? How much does that matter? You think about a running site or a running article where someone like Nike or Reebok would want to advertise. How much does that play in the agent space? At the same time, is there a scenario, maybe this is already in play, is there ever a
possibility that the human advertising and the agent advertising might be different? Give me how that alignment plays out in these scenarios.
[19:47] Mark Howard
We believe that TIME itself is the qualifier in terms of contextually appropriate. We think that agents are coming to TIME in disproportionately high volume because of the authority, because of the way that we still today put all of our editorial through a rigorous process of editors and fact-checking. We focus not on volume but on original storytelling. We believe that provenance is critical, and that if we can be the original source of information that is then either covered by other media outlets or becomes part of the ongoing story, there’s huge value in that. Of course, accuracy is critical, and our ability to continue to use this rigorous publishing process is critical to the core concept. Nothing has changed there.
And that is at the core of what even enables all of this to happen: the great journalism that is taking place every single day and is being published through all of our different channels. But with that, we do believe that TIME itself is what makes it contextually appropriate for these brand ads. As you would imagine, the conversation naturally gravitates to contextually being aligned with the content on the page. With Mobian, we built out that same type of targeting for agent ads that we offer for display ads to people. You can mirror that same targeting capability with agent ads served across these markdown pages. So we do see, and we’ve created, the targeting. You could do it at a section level, a topic level, a persona level,
on the home page. You can do all the same things that advertisers are familiar with with display; you can do with these agent ads now. We’re not so sure that that necessarily matters, because what we do know is that the agents, when they come and they scrape a page, whether it’s the homepage or the Creators hub or a specific article like the one that we put out today about OpenAI, we know that the agents are scraping the page and they’re getting that information. We know that they’re actually reading the agent ad and that they’re able to decipher the ad from the content, and they’re able to store all of that information in their information layer for either real-time presentation or future retrieval. One thing that we’ve observed with some of our partners is
we know that these agents are storing this information that they scrape. And one of the proof points we have for that is that when we look at the referral traffic from the various AI platforms, which of course is relatively small but highly valuable, we do see, through our own GEO and AEO measurement of TIME content being cited on the different AI platforms, oftentimes the most heavily referred articles will have more referrals than they have AI citations, which just isn’t possible. You would need to be cited significantly more times, assuming that maybe on a good day one percent of AI citations result in referral traffic. Even one percent may be potentially high. So you look at your referral traffic, you look at what the
AEO/GEO visibility for TIME content is showing in terms of the number of citations, and we just know that you can’t have more referrals than you have citations. So clearly they’re storing the information. They’re able to retrieve it when appropriate, when needed later on for future queries. And so that storage in that information layer, the ability to be retrieved and displayed later, becomes highly valuable. We actually believe that because of that dynamic,
these agent ad impressions are even more valuable than a single display impression served on an HTML page to a person.
[24:10] Pete Pachal
That’s interesting. I don’t think people even realize it works that way. The way you’re describing it, I don’t know if it’s called caching or whatever, but they generally think of the retrieval as, it’s done and poof, right? But it sounds like it doesn’t work that way. It’s stored somewhere and can inform future referrals. Sorry, I’m just thinking this through in real time as you’re explaining it to me, but you’ve already had more time with it.
[24:43] Mark Howard
Sponsored. Journalists, PR pros and communicators: the fall cohort of AI for Media starts October 13, six live Tuesday sessions with Pete Pachal plus two 1:1 coaching calls. Code AISEARCH500 takes $500 off the $1,500 price for anyone who found the course through AI search, a bigger discount than is offered anywhere else.
Yeah. That’s exactly right. And that’s part of, we study this, we look at it every day. We meet with our partners weekly and monthly. That was a discovery that ScalePost, who we’ve had on the site since the beginning, who we’re really using for a lot of our own AEO and our own AI visibility metrics, looking at the citations of our content, looking at how many times our content was presented and what content was presented and what topics they were and what companies were part of those citations. That’s where this discovery was made about the mismatch between our most highly referred articles and the number of citations. And that disconnect is exactly that proof point of: they clearly are able to retain this information and present it back when appropriate.
[25:37] Pete Pachal
I guess one of the things that comes to mind, maybe it’s related, maybe not, in my own studying of this, is that first seven-day window where most of your citations happen. Off the top of my head, I would just reinforce the importance of that. Because if it’s being stored and creating more of this return traffic, especially in a competitive environment, if you’re competing on the same story with other publications, ensuring that you get that early
retrieval and storage would be even more crucial given this. I don’t know.
[26:13] Mark Howard
Yeah. And it’s a data point, but it’s not the whole story here. What’s actually interesting is when we launch any of our franchises or our lists, that’s very original information, that is IP that we own. And in most cases, these are franchises that we’ve had for a couple of years, or in some cases decades. We see that there is that spike, like what happened with Creators, and then it followed 100 Sports, and now we’ve got TIME100 AI. And so you do get those spikes around those content launches. But actually, what we’re seeing are a tremendous amount of agent crawls of our archives, meaning anything over a year old. And so we do have this tremendous long tail. What we see is a lot of traffic comes to homepage, hubs, and index pages, and then you’ve got this
very, very long tail of traffic to different articles. The articles actually make up the majority of the pages that are being crawled by the agents. And actually the vast majority of those are happening on archive pages, so meaning a year-plus older. So that archive is hugely, hugely valuable in this scenario. And clearly we believe that our authority, the trusted nature of our content, and being the source, the prominence that you can go back to so much of this information that was originally reported on TIME, is the reason for that.
[27:52] Pete Pachal
Nice. So we’re almost a half hour in, and I feel like I need to get to the thing that we were teasing fifteen minutes ago or so, which is that it’s been a month, right, since you launched the agent ads. And presumably now you have some real data points. What do you know now that you didn’t know when you launched?
[28:11] Mark Howard
The big thing, of course, everybody jumps right to measurement. How do we know that this is going to work? And the reality is, like anything that’s launched in terms of advertising, and particularly digital advertising, you have to start somewhere. And so the way that we’re starting, and Mobian designed this great research structure, is really around two things. One is causal ad lift. So it’s really, when prompted versus when not prompted, what are the outcomes of various prompt queries, and how do the various answer platforms respond? So we can run that same kind of test scenario for all of our advertisers. And we start seven days before launch so that we can establish a baseline before the agent ads are even live, so that we’ve got a baseline for all of this research.
And then the other is model impact. So of course, that would be the holy grail here for any brand: if we were able to prove that this, being one of their techniques for how they’re addressing AI visibility, was able to move the needle in any of the different models in terms of how they show up, again with that premise of visibility, favorability, and accuracy. So all of this is being tracked. Right now, of course, being that we’re live with our first two partners, we are going to launch with several others here throughout the remainder of the year. Having more of this data over a period of time is gonna be able to enable us to really drill in more specifically, but it is a robust measurement infrastructure around those two reporting frameworks that become the core
of what we’re able to report on and share. We can show brands pre-launch how they’re showing up. And then it’s this concept of over a period of time. Now, the other important variable here is we believe that with agent ads, just like with display ads, that volume and recency do matter. That even though the AI is still caching, or able to retrieve that information, that as brands’ messaging in market changes, as the products that they’re advertising in market change, it’s going to be critically important for them to continue to keep that updated in these agent ads and in front of the different models. We also know that with the models, the more that they actually do
learn on the same information, the more it captures that information, the better it becomes informed on that information. I think one of the things that’s really important here is actually understanding the construct of the ads themselves. So the agent ads are designed in a couple of different ways. One, we take a URL. In the case of Ally and the Creators sponsorship, if you click on the display ad, it takes you to a page. That page becomes a critical piece of the messaging. The second is you can go to their homepage, or any brand, their homepage or any other pages that really define who they are and what they’re trying to convey to the world about their value prop or their brand value.
So, what I think is important is to understand how the agent ads are built. One of the things that we think about is, from the homepage of any particular brand, the first way to think about whether they are focused on this is if you add /llms.txt to their homepage URL to see if they’ve already created a markdown version of the page. How they constructed their markdown version is very telling in terms of the kind of corporate information that they’re looking to convey and that they want the agents that are coming to their site to actually accurately reflect. The other is looking at what is the click-through URL of any ads that they’re currently running, because clearly that’s what they’re in market trying to talk about. So we take both of those URLs and then we put them into an agent ad builder that Mobian built for us. It’s actually
quite efficient. We do it through ChatGPT. And we just simply plug into the Mobian MCP. We ask ChatGPT to build us an agent ad and to create a demo page that we can display it on, a version of a TIME article, so that we can actually show the brand: one, when we build this agent ad, what are the brand facts about the company that are going to be displayed? Two, what are any mission statements or statements about the company? And then three, building an FAQ to specifically address what they’re trying to convey about themselves or their products or whatever they happen to be in the market. And that FAQ is where we can focus on that gap and where those challenges exist. So then what gets created is a very structured data ad that we then
[33:25] Pete Pachal
Right. Closing that gap that we talked about earlier. Yeah.
[33:42] Mark Howard
use to place on that markdown page. Clearly label the sponsored content going into it. And then there’s an exit tag that identifies that the ad has ended. Also, right before the end, we have the sources of all of the facts that are in that ad so that the agent is able to quickly verify or click through to see that. Those brand-verified facts, just like the name suggests, the brand has to verify that everything in fact is accurate before that ad can actually go live. We then take that agent ad in that structured data format, and then we place that onto the TIME article, again labeling where it starts, labeling where it ends, and clearly labeled as sponsored content.
[34:34] Pete Pachal
So the next logical question is, as that’s ingested, how good are the models at understanding all the labels, the disclosures, and then even if they do, how good are they at displaying them? Right? Obviously people, as we both know, they like their editorial, they’re tolerant of ads, they simply want them labeled so they know what they’re looking at. And so adding this generative AI layer between what you’re putting out clearly and whatever the reader sees at the end seems to introduce a bit of a wild card here. So what have you seen so far in terms of how those systems are interpreting all this? And, forget the numbers for a minute, is it being displayed properly and to the satisfaction of you and the advertiser?
[35:27] Mark Howard
We actually believe that, number one, these AI agents are very smart, of course. Two, right now the way that they are going out to the web, they’re mainly going to publisher websites and going to the HTML version of the page. If you look at the actual code of that page, oftentimes on a publisher article page you’ll have to sift through 30, 50, 100 different mentions of advertisement and sponsorship and all the other code on the page, the JavaScript that the agent is gonna have to sift through in order to get to just the content that they actually care about. You can do that comparison. On our pages it could be anywhere from 30 to 50 times they would have to sift through all of that information to get to the actual content. If you look at the markdown page, we’ve got one clearly labeled
section that signals to the AI agent exactly where the ad is. And you can test it. You can go into ChatGPT or Claude or Perplexity and ask it specifically, when they go to a page on TIME, are they able to see the Ally or the Project Management Institute ad, and what does it say? And it will come back, and each of them have their own formats with how they speak back to you, but all of them very much can identify that the ad is there, that it’s sponsored content, and give you information about what was in there. So they are absolutely able to determine the difference. We think it’s a way cleaner version, where we’re only giving one ad, versus the way that they have to actually go through an HTML page currently and determine what is ad content versus what is editorial content.
[37:20] Pete Pachal
So I guess this gets at the benefit, and maybe I’m interpreting something wrong, to the benefit of the advertiser, that is. Which is to say, if it’s even easier to understand what’s ad and what’s editorial for these, and if someone’s asking about the editorial, then it’s probably trivial for the thing to just throw away the ad. So what is the ad doing? Is it only being used when, am I interpreting this wrong, in other words? Is it the query about the brand that is thus lifted with the editorial that then is the thing? And maybe that’s the disconnect that I think people have from traditional advertising, which is dependent on someone being interested in the editorial. Now it sounds like this is more disconnected, and the lift is more of a brand lift that’s based on, as you’ve described, the authority that TIME has and the long tail and being able to just generally inform AI answers and
being crawled a lot.
[38:20] Mark Howard
That’s right. Remember, what we’re really talking about with brands is their AI visibility. And that is something that they have to address on a number of fronts. They have to think about their own and operated content. They have to think about earned media. They obviously are trying desperately to figure out what are the correlations between the lessons of the last 20 years with SEO to AEO. And it appears today that it’s not. It’s a very different
and it’s a different art form for trying to figure that out. What we’re bringing is an opportunity for another touch point, another surface in a credible environment with a high volume of AI agent scrapes and crawls happening, to get that information, the brand-verified facts, properly into the knowledge layer of the different AIs. And if we can continue to do that, we do believe, based off of this reporting, both the causal and the model impact, that over time we’re going to be able to produce a positive outcome for them around that challenge of AI visibility. But tied very directly, as we discussed today, to where the gaps are between what they advertise about themselves
and how can we, as a platform that is now optimized very much for AI crawling and information retrieval, we do know that when the agents come, it requires far fewer tokens to consume a markdown page. It was actually discussed on one of your podcasts previously. I thought that was a fantastic, that actually was a huge validation point, that entire podcast, for us about this model and what we’ve been developing. And so we believe that, while it’s unproven at the moment, logically we believe that if it requires fewer tokens to get to information that is trusted, that has prominence of the information, that is an environment like TIME where we are the contextually appropriate place for relevant brand facts to also be part of what we’re offering to the AI
agents, that there is a big opportunity here. And to date, we, as mentioned, can prove that the AI can determine the difference between the ad and the content. We can prove that they can surface that information when appropriate to answer a query. And we can prove that over time, when prompted versus prompted to look for the information, in the causal study, that we can actually show an uplift in the results that come back from the AI. So again, this is very, very early. This is version one of what will clearly evolve over the course of the next several months and a couple of years. But as more and more of the web is agents operating on behalf of people, then this idea that we need to be treating that audience as a highly valuable audience
is clear. And so starting that journey with where we are, launching on July 28th, being one month in, but seeing just the evolution of the conversations and the measurement and what we think the opportunity is has been tremendous in the last month. It has definitely not been a dog days of summer August for us. And in fact, we’ve got a couple of other ideas in terms of how to evolve this product on the point that you made, that is the domain authority of TIME, is the credibility of TIME to the AI agents, and our ability to surface brand-verified facts in a format that’s digestible and easily consumable to the AI in a way in which it can help make the AI actually smarter about these brands and their products. And probably more importantly, can help make the AI more accurate
in terms of being able to speak about their products or offerings. Accuracy is gonna be the name of the game for the AI platforms as they continue to compete with each other for consumers. And so we believe that by providing them a trusted source of this structured information, we can play a big role.
[42:48] Pete Pachal
That’s really interesting. That invites such a host of other questions, because I think one of the things that people have a tough time with is that it’s a huge opportunity, but the shape of that opportunity is very different. And there’s so many different things that we don’t know, but we are starting to learn the more we do it. I feel like we just started the conversation and I already have to wrap it up. But I really have to ask about Perplexity. So they’ve been a TIME partner. They’ve called these units deceptive. They said they were going to block it from influencing the Perplexity index. Have they spoken to you directly? Are you still partners? How is their interpretation differing from yours? And can there be a meeting of the minds here?
[43:39] Mark Howard
We did meet with Perplexity, and we actually walked them through the entire process and everything that we’ve developed. We showed them that example of looking at the source code on an HTML page versus looking at the markdown page. We stand by our position that not only is this not deceptive, but it is the most transparent and cleanest way to present to an AI crawler the information. We truly believe that brands need to get, whether it’s their brand facts or their promotions, in front of the LLMs in a way that they can accurately represent them when appropriate based off of any kind of query, that there’s huge value in that. We’ve spoken to a couple of the other AI companies as well. We’ve shown them what we’re doing. We do not, and don’t intend to, operate deceptively at all. We are hugely transparent about it.
We are out there publicly talking about it, to you and many others. The information is all available. If you just simply look at the markdown or the machine-readable version of our page, everything is there for anybody to be able to see. So that is an ongoing conversation with them. We believe that we will continue to add value to them in terms of bringing these brand facts to the surface and making them easily accessible for them. And I can tell you, as of today, we have seen no dip in the volume of Perplexity agents and traffic crawling our site. So you can go into Perplexity, you can ask it questions about whether they can see Ally or Project Management Institute on TIME pages, and it can, and they continue to be able to surface that information. So we feel comfortable that with what we’re doing and how we’re approaching it, the transparency and the openness with all of our partners
and the marketplace in general, that this is only a value add for everybody.
[45:42] Pete Pachal
OpenAI, also a partner. They’re clearly building out their own ad infrastructure and have introduced ads and are giving a lot of indications they’re gonna lean much harder in that direction. Any talk with them?
[45:53] Mark Howard
Yeah, and we’ve shown them what we’re doing. And, interestingly, they’re creating a version of a display ad to serve to people. We’re creating versions of an agent-readable ad to serve to agents. They have been complimentary to date about our approach. They’re building a huge ads business. I think that they’ve got a different goal in mind than the way that we’re thinking about it. But yeah, we’ve been totally transparent with them. We shared with them what we’re doing, and everybody seems to be highly responsive in terms of the way we’re thinking about it. Complimentary. They are engaged, to the point where I feel good that we have been, like any partner would, fully transparent about what we’re doing, and to date they’ve been complimentary about the approach that we’ve taken.
[46:31] Pete Pachal
Responsive. Good word. So just one other topic question for you, because it touches on something we mentioned at the outset, which is the 70 crawlers that you allow. And I’m curious a little bit on the criteria, because it’s probably not all just people you have licensing deals with, and certainly when you have advertising, like we’ve just been talking about for 40 minutes, the more surfaces you can get it on, the better. That said, there are these companies, many of them, some of them are the bigger ones like Exa,
Parallel, that have these agents, and, I’m not trying to loop them in there, I call them sort of gray market, because I know they’d also work with companies like yours on how they can surface the stuff legitimately. But I’m curious on just what’s a legit crawler, I guess, in your eyes.
[47:44] Mark Howard
Yeah. So that is a huge question mark right now about those companies that you mentioned and how they operate. Right now we are not putting those types of companies on the approved list. They are routed via TollBit to a pay gate where they have the opportunity to pay. As we know, they aren’t doing that. And so whether they’re then circumventing the blocking and still coming in and scraping, some are, some aren’t. That is the cat-and-mouse game that unfortunately all the media is playing with those companies. There will be a bigger question about, in the future, whether there’s value in us serving up agent ads to them and letting them through. Right now we’re very controlled over how we build the allow list, who is on it.
It’s more focused. There is definitely a potential in the future that perhaps, if they’re going to continue their practices, that we do want our agent ads to be discoverable and scraped as part of that experience. To date, that’s not part of the strategy, but it is definitely something that is being discussed and could potentially change.
[49:02] Pete Pachal
So last question. What is one thing you are concerned about in this AI future that you’re mapping out here for the publishing world? And what’s a thing you’re hopeful about that could happen over the next six months, two years?
[49:17] Mark Howard
It’s tied to the same answer. It’s pay-per-crawl, pay-per-use. I still really want to continue to participate in industry working groups and with the organizations, the bigger tech companies, that are focused on building marketplaces. There was a lot of excitement and energy around it. It’s sort of slowed down. Some big companies are staying the course in terms of trying to develop how to build that pay-per-crawl, pay-per-use mechanism. I still remain concerned about whether that will materialize in any short order. But I’m optimistic that there’s really great minds that are working on it with huge resources. And if we can get to a model where so much of the value exchange is based off of that pay-per model, that
there can be a new economic underlay that can help journalism and media companies survive through this explosion of the agentic web. Clearly the TAM on that agentic web, when you think about the projections of what it could be, is massive. And then fractions of a penny off of that volume is still material. I don’t think we’re anywhere near, and that that will be the underlying business model, by any means. In the near future it’s an insignificant revenue line item, but I’m concerned, but I’m also optimistic that that will emerge and it will become a part of how we think about running a media organization in the future.
[51:04] Pete Pachal
Nice. Future, hopefully monetizable. Mark Howard, thanks so much for dropping by, telling us about all the cool stuff you guys are up to at TIME. Love to check in with you again sometime in the future, see how it all played out.
[51:15] Mark Howard
Love to. Thanks, Pete. Thanks for having me again. Take care.
[51:18] Pete Pachal
All right, take care. Thanks for listening to my conversation with Mark Howard, Chief Operating Officer of TIME. Media Copilot is a podcast, so please follow or subscribe to us on the platform you’re listening to us on. If you’re on Spotify or Apple, please leave a five-star review and a note about what you like. And if you’re on YouTube, please like the video and subscribe to the channel. All those things really do help the show because they put the podcast in front of more people. The Media Copilot is produced by Musso Media.
Look for their link in the show notes, and don’t forget to subscribe to the Media Copilot newsletter as well. Thanks again for being here. Seeing you in the future.
