Skip to content
← Back to Adoption

Google AI Overviews cut clicks to other sites by 39.8% where shown, a randomized test found

The 40% drop in the headlines applies to searches that show an Overview. Across all searches the authors estimate about 18.5%.

0.37 vs 0.62Clicks to other sites per search, with vs without an AI Overview

Saharsh Agarwal (Indian School of Business) and Ananya Sen (Carnegie Mellon) ran a preregistered randomized experiment on US desktop Chrome users. A browser extension either showed or hid Google’s AI Overviews, the AI-written answer at the top of the results page. It tracked two weeks of real searches per person between 7 January and 10 February 2026. Of 1,065 participants, 770 were in the two groups compared (396 shown Overviews, 374 with them hidden).

On searches where an Overview was due to appear, people who saw it made 0.37 clicks per search to outside websites. People who did not see it made 0.62. That is a 39.8% reduction. With the Overview, 73% of those searches ended with no click to another site, against 54% without it. Ad clicks and the number of searches did not change. Overviews appeared on about 41% of searches, which is how the authors get to about 18.5% fewer organic clicks across all searches.

A separate preregistered experiment by Stephanie Wang and colleagues (1,100 people) found that removing AI Overviews and AI Mode raised click-through rates to publishers. Its abstract gives no figures. In July 2025 Pew Research Center, using the browsing of 900 US adults, reported clicks on a result link in 8% of visits with an AI summary and 15% without. In August 2025 Google’s Liz Reid wrote that organic click volume was “relatively stable year-over-year” and that average click quality had risen.

What it means

How much a site loses depends on how many of its visitors arrive through searches that show an Overview. Sites that answer factual questions are the most exposed, since informational searches were where the effect was clear.

Caveats

  • It is a working paper, not peer reviewed. We read the 17 June 2026 version, the date the draft carries; SSRN itself blocked our access.
  • Participants were recruited through Prolific, a paid-participant platform. The study covers US desktop Chrome users for two weeks, so it is not a sample of all searchers.
  • The effect was significant for informational searches, which were 71% of searches. The navigational and transactional samples were too small to read.
  • Google disputes third-party reports of steep declines, calling them methodologically flawed. It reports stable total clicks and says traffic is shifting between sites. An experiment that hides the feature does not test that claim about the whole market.
  • Pew’s figures are observational, so they cannot separate the Overview’s effect from the kind of search that triggers one.
Suggest a correction