Google AI Overviews Cut Publisher Clicks 39.8%, Study Finds

The first causal estimate

The argument over whether Google's AI Overviews take traffic from publishers has mostly relied on before-and-after comparisons. Those comparisons can be confounded by algorithm updates, seasonality, and changing user habits. A new randomized field experiment offers more direct evidence, and it finds a large effect.

Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University's Heinz College tracked 1,065 desktop Chrome users through a browser extension. They then measured what happened when an AI Overview appeared and when it did not. 1 When an Overview was shown, outbound organic clicks fell by 39.8% and zero-click searches rose by 34.5%. 1 The paper appeared on the Social Science Research Network on April 3, 2026, with a revision dated June 17, 2026. The authors describe it as the first causal evidence on the question. 1 Earlier work was observational. This study is not, and that is why the result carries extra weight. 1

Citations don't fill the gap

Google's usual defense is that AI Overviews link to their sources, so publishers still benefit. The experiment tested that claim directly. Digital Content Next, the publisher trade group, says the source citations inside Overviews recover only a small share of the lost traffic. 2 In practice, being cited in the AI summary does not make up for losing the click that a traditional blue link would have earned.

Two other findings shape how the result should be read. Clicks on sponsored results stayed flat when an Overview appeared. 1 Users were also no more satisfied with their searches. 1 Taken together, these suggest the Overviews moved attention away from organic results while leaving Google's ad clicks intact, and without a measurable improvement for users. That combination is what makes the study awkward for Google. If users were clearly better served, the lost traffic could be framed as a reasonable trade-off. Here the data show no such benefit to point to. The ppc.land write-up notes that this raises new regulatory questions. 1

Different readings of the same study

The two outlets covering the research largely agree on the numbers but frame them differently. ppc.land emphasizes the experimental method, the flat sponsored clicks, and the regulatory angle. 1 Digital Content Next, which represents publishers, focuses on the business conclusion. In its view, search is becoming a weaker channel for discovery and referral traffic, and publishers need new ways to grow their audiences. 2 The two framings do not conflict. One is about what the evidence shows, the other about what publishers should do next.

The usual caveats apply, and they are my analysis rather than the authors'. The sample covered desktop Chrome users who agreed to install a tracking extension. Mobile behavior, where many searches now happen, may differ. A 39.8% drop measured on queries that triggered an Overview is also not the same as a 39.8% drop in a publisher's total search traffic. The overall effect depends on how often Overviews appear for that publisher's queries.

Google keeps expanding AI answers

That last point matters because Overviews are appearing more often and getting larger. Yext's running log of search changes reports that Google has been expanding AI Overviews into full, AI Mode-style responses by default for more queries. The result is fewer clicks to organic and local results. 3 Google has also launched AI Mode "info monitoring" worldwide. The feature lets an AI agent keep checking information on a user's behalf, which turns search from a single query into an ongoing service. 3

Google is also testing ways to make money from the new format. Text link ads are being tested in AI Mode. 3 Google is also running an early-stage test of paying publishers whose content is used in AI Mode, AI Overviews, and Gemini. Yext calls this the first concrete monetization path for publishers whose work powers AI search. 3

What it adds up to

The overall picture is consistent. One controlled experiment finds AI Overviews cut organic clicks by about two-fifths. Citations recover little of that, users gain nothing measurable, and ad clicks are unaffected. 12 Meanwhile, Google is putting AI answers in front of more queries and starting to sell ads inside them. 3

The publisher payment test is a tacit acknowledgment that the old deal, where publishers supplied content and Google sent traffic back, is breaking down. Whether those payments come close to offsetting a 40% loss in clicks is unknown, and the test is still early. 3 For now, publishers should treat search referrals as a shrinking source of traffic, not a dependable foundation. Regulators also now have causal evidence of the harm, rather than only industry complaints.