Google’s AI Overviews Cut Outbound Clicks by 40%, Experiment Finds

When Google shows an AI summary at the top of its results, far fewer people click through to actual websites. That was the suspicion for two years. Now a randomized experiment has measured it: showing an AI Overview cuts outbound organic clicks by about 40%.
The study, published as a working paper by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon, is the strongest evidence yet that AI-generated search summaries are diverting traffic from publishers — and the analysis behind it was published on November 18.
How the experiment worked
The researchers gave US desktop Chrome users a browser extension that randomly assigned each person to one of two groups. One group saw Google search as usual. The other had AI Overviews hidden. Then they watched two weeks of real searches, in January and February 2026, and counted what happened. With 1,065 participants, it is a randomized controlled trial — the most reliable kind of study for separating cause from coincidence.
On searches that would ordinarily trigger an Overview, the group that could see them averaged 0.37 clicks to outside websites per search, against 0.62 for the group that could not. The researchers report that as a 39.8% reduction. Searches ending with no outbound click at all rose from 54% to 73% when Overviews were present.
A second, separate preregistered experiment by Wang and colleagues, with 1,100 people, found the same direction: removing AI Overviews and AI Mode raised clicks to publishers. Two experiments with different designs, one conclusion.
The numbers that matter
The headline figure is 40%, but the details sharpen the picture. The loss was concentrated in informational searches — the research questions people ask before they buy anything. The effect was statistically significant there; the samples for navigational and transactional searches were too small to settle anything.
Across all searches, including ones that never show an Overview, the authors estimate the feature costs websites about 18.5% of the organic clicks they get from Google. And that share will grow: DataForSEO’s panel shows AI Overviews on 61.43% of tracked US keywords in August 2026, up from 5.93% in September 2024. More searches get summaries every month, so the traffic loss compounds.
Other data points the same way. Chartbeat data for the Reuters Institute showed Google organic search traffic to more than 2,500 news sites falling by a third globally in a year, with small publishers losing 60% of their search referrals over two years. The Reuters Institute is careful to say AI summaries are not the only cause — changes in search and in audience behaviour matter too — but the experiment now shows the summaries alone cause a large part of it.
Why ads didn’t budge
Here is the detail that will raise eyebrows. While organic clicks fell by about 40%, clicks on ads did not change at all. The lost clicks were not low-value ones either: bounce rates, quick exits and time on site showed no measurable difference between the two groups. People who clicked through from a search with an Overview behaved just like people who clicked without one — there were simply far fewer of them.
The ad result narrows the argument in an important way. It does not show search becoming too expensive for advertisers. It shows search becoming less dependable as a route outwards for everyone who is not paying Google. The visit that paid for journalism is being replaced by an answer that pays for nothing, while the paid slots beside it keep working exactly as before.
What publishers can actually do
Google is now paying a small group of publishers through an invitation-only AI contribution pilot, reportedly covering around 100 publishers, many of them small and mid-sized. That is not a plan for most websites. Search Console’s generative AI performance report, rolled out to all sites in August 2026, shows how often your pages appear in AI Overviews and AI Mode — but it does not show clicks or earnings.
The industry’s answer so far has been GEO, or generative engine optimization: writing so AI summaries cite you. The evidence that it compensates for the loss is thin. A top search ranking may no longer deliver the same traffic when a generated answer sits above the links, and being cited inside the answer does not pay the bills the way a visit used to.
The honest takeaway is that the old bargain — write good content, rank well, get the click — is breaking. Publishers need readers who come to them directly: newsletters, loyal audiences, subscriptions, and content worth returning for. Google’s AI answers are not going away, and this experiment shows exactly what they cost the open web.
The study has real limits, and they should be stated plainly. It is a working paper, not yet peer-reviewed. It covered desktop searches only, and it observed people for just two weeks. The numbers are mainly US English, and Google’s products change quickly. Even with those limits, a randomized trial is the best evidence anyone has produced — and it points one way.
