The hard truth: AI kills search traffic
Confirming what we all knew, Tollbit released one of the first definitive studies on AI search referrals, and it's not a pretty picture for publishers.
AI startup Tollbit dropped some hot new data this week, attaching some numbers to just how much search traffic drops when your content is linked in AI search vs. regular search. It's pretty dramatic, and it's yet another reminder that media companies need to adapt to the AI future sooner than later.
More on that in a minute, but first a quick note that you have just over a week to register for the spring cohort of AI for PR & Media Professionals, the six-week AI training course that I offer in partnership with my friends at The Upgrade. The first cohort was tremendously successful, and all the material is battle-hardened and ready. Need to know more? Scroll down 👇
A MESSAGE FROM THE MEDIA COPILOT & THE UPGRADE
Leverage AI and Stay Ahead in PR
Generative AI is now essential for PR professionals. That's not me saying it — it's according Muck Rack's State of AI in PR report for 2025, which debuted just this week. The report reveals 75% of PR professionals are now using generative AI, with 93% saying it speeds up their work. Yet, only 38% of companies have policies guiding AI usage.
💡 AI for PR & Media Professionals
The guidance you need is here: a six-week live online course starting March 18, designed to give PR pros, media strategists, and content creators the edge they need in an AI-driven landscape.
What You’ll Learn:
AI-Enhanced Content Creation: Automate press releases, refine pitches, and master audience insights.
Time-Saving Workflow Automations: Automate repetitive tasks to free up time for strategic decision-making.
Campaign Optimization: Use AI analytics to measure ROI and strategize for success.
🎓 Taught by Industry Leaders
Join Pete Pachal (Founder, The Media Copilot), Peter Bittner (Lecturer, UC Berkeley) and Kris Krüg (AI consultant for creatives) for expert guidance, personalized coaching, and insights into the tools shaping the future of PR.
The brutal numbers behind AI search
When Google first announced its AI Overviews, we wrote that the move was a "death blow to publishers," since it would mean more people reading summaries of articles than the articles themselves. Subsequent information may have seemed to contradict that assertion: Search traffic has been relatively stable for news websites, and Google itself came out last fall to declare that Overviews actually send higher-quality visitors to sites, meaning they were likely to linger (and perhaps even buy something). If you squinted, you might start thinking this AI future might actually turn out OK.
Well, you can probably toss all that optimism into the trash in the wake of a new study from Tollbit, a startup that helps media companies protect content from AI crawlers. According to the report, the amount of referral traffic from AI search engines is drastically less than what publishers typically garner from Google search. Data from the sites in Tollbit's publisher network, which includes Time and AdWeek, shows that click-through rates from AI search is 91% less than from traditional search.
When looking at AI chatbots (as opposed to AI search engines), the gap is even worse: assuming they can even link out, chatbot click-through is 96% lower than search.
Google giveth, AI taketh away
So… time to panic? Hold up for a second. AI search still accounts for a fraction of search activity. On the high end, ChatGPT only accounts for a small fraction of online search, and Perplexity — whose whole business is AI search — is even smaller. Google does AI search, too, of course, but it hasn't yet begun using its AI Overview summaries on news stories.
However, that day seems inevitable. AI services are beginning to chip away at Google's market dominance, and Google has been tiptoeing toward AI-summarized news, rolling out news summaries for Google TV at CES 2025. In the meantime, it continues to expand and enhance AI Overviews, spreading them to Google users who aren't logged in, and beginning a broader rollout of the more advanced "AI mode" for search this week. The future of Google search includes more AI, not less.
If there's a silver lining in the Tollbit report, it's that we now have a different perspective on the value of AI licensed content. Up until now, most assessments have looked at AI content licensing from the point of view of the AI company. A good example is Atlantic CEO Nick Thompson's back-of-the-envelope math for signing a deal with OpenAI, which he revealed at Reuters Next last fall:
Start with the value of OpenAI (at the time, about $250 billion).
Estimate what portion of that was the training data (about $75 billion).
Estimate how much of that was media content (say, $5 billion).
Work out what percentage of that is attributable to The Atlantic (somewhere in the low millions).
Finally, determine what portion should belong to The Atlantic (the industry seems to be settling on 50%, but it's been suggested it should be higher) and you have your number.
The publisher perspective
That's all well and good for calculating the value of publisher content to OpenAI, but with the Tollbit data, we now have a better picture of the other side of the value equation: what AI search might cost publishers. That lets us reverse the telescope.
Imagine a media company with $10 million in ad revenue generated by impressions on its website (for simplicity's sake, we'll use round numbers). Now, estimate the portion of traffic from Google referrals — say, 40%, or $4 million.
The Tollbit report specifies the average amount of referral traffic from AI engines. It's been steadily growing for the past year, but it was about 2% in November 2024, so let's go with that. Remember that click-through from AI search is 91% lower than traditional search. That comes from this chart, which shows the raw CTR rates: Google search is 8.63% while AI search lingers down at 0.74% (regular AI chatbots like ChatGPT have a meager 0.33% CTR).
So, if 2% of your traffic is AI search, you can project what the actually Google traffic would have been easily: 2% x 8.63/0.74 = 23.3%
So, if you take the $10 million total revenue from ads, the revenue lost from AI search comes out to $2.3 million, which sounds about right when compared to the $4 million in Google referral revenue. Of course, let's not forget to subtract the people who actually did click through, 2% of our impressions, which is worth about $200k. But let's be generous (those eyeballs are "high quality," after all) and round up to $300k, which gives us an even $2 million in lost revenue.
To make everything cleaner, I only considered AI search, not AI chat. That would complicate the calculations, and it's not clear from the Tollbit data how much AI usage is search vs. chat. And that's just for starters — there's obviously a whole host of factors on both the publisher and AI sides of things that would throw the numbers even more.
Still, as a back-of-the-envelope calculation, it gives a loose guide for the potential revenue that might be lost through AI search. That in turn is a pretty good starting point for media companies looking for fair value for their content as they pursue licensing deals with AI companies or intermediaries like Tollbit, ProRata and others.
Looking at the landscape of AI search, and the looming threat of Google making further moves within it, publishers have a right to be worried. At least now we know a little more about exactly how much.
The Chatbox
All the AI news that matters to media*
AI plays referee for LA Times opinion section
The LA Times launched AI-powered bias ratings, called Viewpoints, for many of its opinion articles this week. The move is part of a broader revamp of the paper's opinion section, now renamed Voices, apparently instigated by billionaire owner Patrick Soon-Shiong. The bias ratings are powered by the AI news app Particle, identifying where on the political spectrum the article falls (such as center right), as explained in an article on Medium. In addition, some Voices articles include AI-generated "Perspectives" — summaries of alternative points of view, with links. That feature is powered by Perplexity.
Notably, both the Viewpoints and Perspectives appear at the bottom of articles, and the user needs to click on them to see what they are. This seems to be a move to shake out any problems before giving them more prominence. And there do seem to be problems: A Nieman Lab deep dive found some oddities in the sourcing of Perspectives, and a piece on the Ku Klux Klan of the 1920s produced an AI-summarized counterpoint that was quickly removed after a BlueSky user pointed it out. This isn't the first time AI is acting as a tool to identify and mitigate media bias, however, and it certainly won't be the last.
Patch turns community coverage dial to 30,000
The solution to revitalizing local news that comes up the most often is newsletters, mostly because they're just so damn easy to spin up. Patch, the news platform that specializes in hyperlocal coverage, is multiplying that ease by the power of AI to expand from just over 1,000 communities served by its 85-person editorial staff to more than 30,000. Those "unmanned" newsletters aren't as meaty as the human-produced variety, CEO Warren St. John tells Axios, but they do include unique content relevant to the community, such as local weather and first-responder updates. And where there's sufficient interest, Patch may hire an actual reporter to do the newsletter, as it did in Georgia. That might be the first confirmed case of a human taking the job of an AI.
Google's AI Overviews go next-level
The new AI Mode in Google search, currently only available to select people who opt in, enables complex, multi-part questions. It's a clear attempt to make Google's user experience more like Perplexity or ChatGPT. Under the hood, the feature uses a "query fan-out" technique, which concurrently searches multiple data sources to create a single coherent response. Google VP Robby Stein tells TechCrunch that user queries are twice the length of traditional search terms and they follow up 25% of the time, suggesting a more conversational relationship with search. Left unanswered: whether Google will ever feel it needs to license any of this content.
Writing the playbook for AI search rankings
Speaking of search, a startup called Scrunch AI is hyperfocused on the AI-powered variety, and decoding how to "win" at what counts as SEO for chatbots. The market opportunity is clear: helping companies adapt to a world where AI bots visit websites more than humans. The company recently secured $4 million in seed funding, according to TechCrunch, and provides tools for enterprises to audit and optimize how their content appears in AI search results. With 25 customers already signed up, including major enterprises like Lenovo, Scrunch AI's success signals a growing recognition that brands are no longer defined solely by their own messaging but increasingly by how they appear in AI-generated summaries. (AI-assisted)
Sesame makes machines sound authentically messy
Sesame's new AI voice models could be a watershed moment for podcast creators. The technology's ability to include human-like imperfections — breathing, stumbling over words, and natural conversational flow — signals how quickly the distinction between synthetic and authentic voice content is disappearing, based on a thorough look from Ars Technica. These tools could dramatically reduce production costs for podcasts, customer service, and audio content, while simultaneously creating new verification challenges as voice becomes increasingly spoofable. The technology can even roleplay angry characters (which ChatGPT refuses to do). Just as media companies developed standards for AI-generated text and images, they may soon need to create frameworks for how synthetic voice fits into content strategy. (AI-assisted)
*Some items are AI-assisted. For more on what this means, see this note.