TLDR
A randomised field experiment published on SSRN found Google AI Overviews reduce outbound organic clicks by 39.8% and lift zero-click searches by 34.5%. The clicks that disappeared were not low-quality: bounce rate, time on site and return-to-search showed no measurable difference between users who saw AI Overviews and those who did not. The findings directly contradict Google VP Liz Reid's August 2025 claim that click volume was stable and average quality was rising. Publishers are losing engaged visitors, not junk traffic.
KEY TAKEAWAYS
What the numbers actually say
A randomised field experiment found Google AI Overviews reduce outbound organic clicks by 39.8% when shown.[1] The same experiment recorded a 34.5% rise in zero-click searches, queries where the user never leaves the results page at all.[1]
Those are large numbers. Nearly four in ten clicks that publishers would otherwise receive are being absorbed at the search page, redirected into an AI-generated summary and never converted into a visit.
How the experiment was built
Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University designed the study as a randomised field experiment, the methodological gold standard for establishing causation rather than correlation.[2] The paper was published on SSRN on 3 April 2026.
Randomised design matters here because it rules out the most common objection to this kind of traffic analysis: that queries receiving AI Overviews are inherently different in type or intent from those that do not. By assigning users to conditions randomly, Agarwal and Sen can attribute observed differences in click behaviour directly to the presence of the AI Overview rather than to the nature of the search.
The quality question
Downstream engagement on publisher websites, measured by bounce rate, time on site and return-to-search, showed no measurable difference between users who clicked through after seeing an AI Overview and users in the control group.[1] That finding does the most damage to the prevailing narrative around AI Overviews.
If the clicks being removed were shallow, users who would bounce immediately, spend seconds on a page and return to Google, you would expect the remaining pool to look meaningfully better on those metrics. The data show nothing of the sort. Visits being lost look, by every measured indicator, identical to the visits that remain.
Survey evidence collected alongside the experiment reinforced the point. Users shown AI Overviews reported no measurable improvement in perceived search quality or ease of finding information compared with users who were not.[1] The feature is not generating a demonstrably better search experience; it is generating fewer clicks.
Google's public position
Google VP and Head of Search Liz Reid addressed publisher traffic concerns directly in August 2025. "Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year," Reid said.[3]
Reid went further on the quality argument. "Additionally, average click quality has increased and we're actually sending slightly more quality clicks to websites than a year ago," Reid said.[3] Agarwal and Sen's experiment, which produces causal evidence rather than aggregate trend data, finds neither condition holds when AI Overviews are the variable under examination.
Google's position rests on year-over-year aggregate comparisons, which blend many signals into a single number. A randomised experiment isolates the effect of one variable. The two methodologies are not directly equivalent, and the causal design has the stronger claim on what AI Overviews specifically do to traffic.
What publishers are losing
The practical implication of the engagement finding is straightforward: publishers are not being relieved of tyre-kickers. Agarwal and Sen's study, the first to provide causal evidence on the question, indicates that visitors disappearing into zero-click searches would have read the content, stayed on the page and not immediately bounced back to Google.[2]
AI Overviews are AI-generated summaries that appear at the top of eligible search results, drawing from web sources to synthesise an answer without requiring the user to visit any of those sources. Google has maintained the feature mainly strips out low-quality clicks. The experiment's engagement data suggest that framing does not hold up.[1]
The study was presented at the Toulouse School of Economics on 2 July 2026, with Agarwal and Sen's SSRN publication dated 3 April 2026.
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
What did the AI Overviews study find?
Who conducted the study and when was it published?
Why does the randomised design matter?
What has Google said about AI Overviews and click traffic?
Takeshi Mori covers startups and technology for Bushletter. He is impatient with hype and interested in how products actually get built.
