It’s been a big week for AI-powered search. Between the official launch of the Perplexity Publishers’ Program and OpenAI previewing SearchGPT, the online world has a better idea of how audiences will use AI search engines, and the mechanisms behind them.
What does this all mean for the king of the magnifying glass, Google? That’s what I dive deep into for today’s newsletter.
Before I get to that, though, I’d like to welcome new readers from McClatchy newspapers, The Washington Post, Australia’s ABC network, Forbes, and Harvard Business. It’s a pleasure to have you here!
Now let me get a quick word about The Media Copilot’s upcoming AI training classes for August…
Summer is the perfect time to enhance your AI skills for a productive September. The Media Copilot offers virtual classes specifically designed for media professionals, including journalists, marketers, and PR experts.
Key features of our classes:
Live, interactive instruction
Comprehensive coverage from basic prompts to specialized tools in a single afternoon
Upcoming classes:
AI Quick Start (Aug. 22):
Duration: 1 hour
Cost: $80 per person
Focus: Essential AI basics and efficient prompting techniques
AI Fundamentals (Sept. 4):
Duration: 3 hours
Cost: $375 with discount code AIHEAT (normally $750)
Focus: Advanced prompting, ethical use of image generators, and a curated toolkit for daily efficiency
Reserve your spot, enhance your skills, and be ready to impress your colleagues.
Google’s AI Weak Spot
Believe it or not, the AI world had some good news for the media this week.
First, OpenAI teased SearchGPT, a search engine that uses the power of AI to produce summaries of what it finds on the web. Yes, ChatGPT could browse websites before, but now there’s attribution: there are prominent links to source material, which is great for publishers, especially those who have signed licensing deals with OpenAI.
Then came the Perplexity Publishers’ Program. The program allows publishers to share in advertising revenue generated on any answers where their source material is used. I had an extended conversation with Perplexity Chief Business Officer Dmitry Shevelenko about it, and he told me the goal is to make the program scalable so any content provider could sign up.
Taken together, the dual announcements may be the precursor of a new AI search economy, which I wrote about on Tuesday. However, if you’re talking about how people search for information online, Google gets a say. And so far it hasn’t said much outside of releasing its own AI-powered search, AI Overviews (née “search generative experience”).
Google clearly sees AI search as the future, but so far its stance leaves content providers out of the conversation. As a trillion-dollar company, Google obviously has an interest in preserving the status quo — at least from a business standpoint — but that’s providing an opening for Perplexity and OpenAI.
At a glance, the approaches of Google’s two competitors look somewhat different. OpenAI is paying straight licensing fees to publishers while Perplexity’s model only pays publishers when their content is used in answers that have advertising.
However, they’re probably more similar than they first appear. According to a source familiar with OpenAI’s publisher negotiations, the deals have two parts: a one-time fee for using archives for training models, and “metered” payments based on usage in search results. The Perplexity agreement is more like the latter. And while OpenAI’s business model so far doesn’t include advertising, many believe that is inevitable, given how commoditized foundational models are becoming.
A Brief History of AI Search
The difference between training data and browsing is important, even if, to the user, the result is the same: a summary of the requested information. In one case the AI effectively “internalizing” the information, and in the other it’s simply reading something online and summarizing it.
In the latter case, linking to the source is considered a best practice. Traditionally, a link alone does not include much of the actual information on that page. Yes, there’s almost always metadata and “cards” that provide some summarized information, but it’s typically a tease and doesn’t reveal key info. And content creators have a good deal of control over how these cards were presented.
AI summarization, which is clearly on the rise, is something different. While not a perfect comparison, the act of taking an article, paraphrasing it, and combining that information with other sources, is really more like syndication than linking.
Syndication, of course, is when a publication makes deals with other websites or platforms to republish their content, usually in their entirety. They do this for a variety of reasons, and compensation can vary widely. Sometimes there’s no money exchanged and the benefit to the publisher is to get exposure to new audiences, but often there’s also money involved, which typically takes the form of a revenue share model.
With the emergence of SearchGPT and Perplexity’s program, the industry seems to be steering toward the perspective that AI summarization is more like syndication. As The New York Times lawsuit progresses, we may even get a court ruling that supports that view.
Between an LLM and a Hard Place
All of this puts Google in a spot that might get increasingly uncomfortable. Google — far and away the leader in search — has pushed full steam ahead on its AI Overviews, which summarize information on the topic being searched above the actual search results.
Here’s how a Google spokesperson described AI Overviews when I asked about the company is positioning them: “With generative AI, we’re continuing our work to expand what Google Search can do and open up new types of questions that we can help with, creating new opportunities for web content to be discovered. AI Overviews are designed to provide a snapshot of relevant information from multiple web pages and include links so that people can click out to explore further and learn more.”
From the statement, it’s apparent that Google sees Overviews as an extension of traditional search, which is all about linking, and not something fundamentally different that might require a new relationship with publishers and content creators.
Google hasn’t revealed what portion of searches provide Overviews, but there’s no question it will continue to grow. Also a factor: Overviews generally don’t appear when the search is about a current news story, but again, that will probably change at some point, too — especially if SearchGPT and Perplexity are increasingly used for that use case.
You can see why Google has made zero moves to create either licensing deals like OpenAI or a publisher program like Perplexity. If they did, it would mean taking a big hit on potential future revenue as AI search becomes the norm — not to mention a de facto concession that AI summarization requires some kind of agreement with content providers.
On the other side, publishers still depend greatly on Google. You can’t just sue them or opt out of being crawled — that would be devastating to the search referrals that websites depend on. Yes, a site could technically adjust its robots.txt protocol to block AI but not search crawlers (and the Google-Extended portal specifically gives publishers a way to opt out of training its AI), but exactly how AI search works continues to evolve.
There’s room here for compromise. As Google looks to bring a generative experience to Google News, the moves made by Perplexity and OpenAI provide a guide. Google has all the talent and resources to make truly great generative news product, and publishers would be extremely motivated to work with Google on it.
AI search is a brand-new industry, and ironically the leader in search is starting on the wrong foot by showing little to no deference to content providers. The company has drawn a hard line against considering an altered business model — threatening to kill its Google News Initiative in response to a California bill that would have forced it to pay publishers — revealing it’s more fearful than visionary. If Google ends up losing the AI search ace, that’ll be the reason.
News of the week is below…
Keep Your SSN Off The Dark Web
Every day, data brokers profit from your sensitive info — phone number, DOB, SSN — selling it to the highest bidder. And who’s buying it? Best case: companies target you with ads. Worst case: scammers and identity thieves.
It’s time you check out Incogni. It scrubs your personal data from the web, confronting the world’s data brokers on your behalf. And unlike other services, Incogni helps remove your sensitive information from all broker types, including those tricky People Search Sites.
Help protect yourself from identity theft, spam calls, and health insurers raising your rates. Plus, just for The Media Copilot readers: Get 55% off Incogni using code COPILOT.
The Chatbox
All the AI news that matters to media
ChatGPT Finds Its Voice: The new voice features for ChatGPT — the ones the company previewed in the spring and received the attention of Scarlett Johansonn’s lawyers — are beginning to roll out to users, the Verge reports, and increased silliness on X confirms this. It should be in wide release by the fall, around the same time the new iPhone is due to come out, but Apple notably will be delaying its Apple Intelligence features, which brings some AI features to Siri and other Apple apps. It’ll be interesting to see who gets there first. If it’s ChatGPT, OpenAI has an opportunity to define what AI on a phone is, at least in the minds of some iPhone users.
So We’re All Just Training on YouTube Then? Samantha Cole at 404 Media got a hold of an internal document from generative video service Runway that shows the company trained its AI on YouTube videos from many major publishers and entertainment companies, including Disney, Sony, The New Yorker, and many others. That’s a no-no according to YouTube’s terms of service, but more to the point: the whole practice of scraping “publicly available” data to train AI models, while a de facto standard in the tech industry, is now in territory that’s very legally dicey. We anxiously await the inevitable court ruling on this.
AI Spreads in Academia: A great deep dive in Nature explores the sudden explosion of the use of AI writing in academic work, and it attacks the key question: Is AI writing the same as plagiarism? If it is, a lot of authors are in trouble, since use of AI as a writing tool has skyrocketed in 2024. Obviously, the answer is “it’s complicated,” since AI (when working correctly) produces original work. But without human-generated content to fuel it, a large language model has nothing to go on. In any case, it’s clear universities will need to adopt policies on the best practices on how to use AI to ensure authors don’t get lazy and overuse the word, “revolutionize.” Call us — we can help with that 🤙🏻
Leave a Reply