Every few weeks a new number arrives claiming to measure exactly how much AI search summaries are costing publishers, and every few weeks that number gets repeated everywhere without anyone checking where it came from. The underlying question, how much click-through traffic AI Overviews and similar summaries are actually removing from the web, is real and worth answering. The answer is messier than any single headline stat lets on.
What the antitrust filing claims
The figure making the rounds right now is that AI Overviews have cut clicks by roughly 58 percent. That number did not come from Google. It traces back to Ahrefs research published in February 2026, which compared click-through rates on 300,000 keywords between December 2023, before AI Overviews rolled out widely, and December 2025. For queries that triggered an AI Overview, Ahrefs found the click-through rate for the top-ranking result fell from 7.3 percent to 1.6 percent, a drop it calculated at about 58 percent after adjusting for broader search trends.
That research took on a second life when Penske Media, which owns Rolling Stone, Variety, and Deadline among other titles, cited it in a February 2026 memorandum opposing Google's motion to dismiss its antitrust suit over AI Overviews. Penske's underlying claim is that Google is using its search monopoly to keep visitors on Google's own page instead of sending them to the sites Google indexed in the first place. The 58 percent figure supports that argument, which is exactly why it is worth remembering it is an allegation built on independent research, not a number Google has confirmed or published itself.
The older, calmer number
Before the lawsuit gave the topic a fresh headline, Digital Content Next, a trade group for premium publishers, ran its own check. In an August 14, 2025 report based on a member survey covering 19 companies, including major newsrooms and entertainment brands, DCN found a median year-over-year decline of about 10 percent in referral traffic from Google Search over an eight-week window. The drop was not even across the board: news brands were down a median of 7 percent, while non-news brands fell further, a median of 14 percent. DCN also noted that declines outnumbered gains by roughly two to one, with the worst individual weeks showing drops as steep as 16 to 17 percent for some publishers.
Ten percent and 58 percent are not measuring the same thing. DCN's number is a broad, self-reported median across many kinds of queries and publishers. The Ahrefs figure isolates the specific case of queries that show an AI Overview, where the effect on the top result is naturally larger. Both can be true at once. Neither should stand in for the other.
Publishers are pulling access, not just complaining
Some of the response has moved past public statements and into robots.txt files. A BuzzStream study of the top 50 news sites in the US and the UK, 100 sites after removing overlap, first published in December 2025 and updated in April 2026, found that 79 percent block at least one bot used to train AI models, and 62 percent specifically block OpenAI's GPTBot. Separately, 71 percent block at least one AI live-search or retrieval bot, with 67 percent blocking PerplexityBot by name.
The catch, which the study is upfront about, is that robots.txt is a request, not a lock. Bots that ignore it face no real penalty unless a publisher enforces the block at the network or CDN level. So the blocking numbers say a lot about publisher sentiment and not as much about how effective the blocks actually are at keeping AI systems out.
Why the measurement is genuinely hard
None of this comes with an agreed-upon ruler. Ahrefs, DCN, and the various crawler studies use different samples, different time windows, and different definitions of a click or a referral. Google has not released its own figures on how AI Overviews affect publisher traffic, so every outside estimate is built from whatever slice of data that group happens to have access to, a search-engine results panel, a set of member analytics accounts, or a crawl of robots.txt files. Correlation and causation also get blurred easily. Referral traffic was already shifting before AI Overviews existed, driven by app-based browsing, social platforms, and Google's own search layout changes. Isolating the AI Overview effect from all of that is possible but not simple, and most of the numbers in circulation do not fully attempt it.
How to read the next headline like this
A specific number attached to AI and traffic loss is usually a compressed version of a longer study, and the compression is where the honesty tends to leak out. Before treating one as settled fact, it helps to ask a short list of questions: is this a measured result or an estimate, who ran it and what is their stake in the outcome, does it describe all search traffic or a narrow slice of it, and is it appearing in a lawsuit filing, where a party has a clear reason to pick the most dramatic supporting figure available. None of that means the underlying trend is fake. Publisher referral traffic from search does appear to be declining, and multiple independent measurements point the same direction. It means the size of that decline depends heavily on which study you are reading and what it was actually built to measure.
This piece follows Encore Editorial's own rule for statistics: trace the number back to where it started before repeating it. Questions go through our contact page.