Google AI Overviews Study Finds Lost Clicks Weren’t Lower Quality

Google Still Hasn't Shown That AI Overviews Only Steal Low-Value Clicks

A revised working paper from researchers Saharsh Agarwal and Ananya Sen finds that when Google's AI Overviews are removed from search results, the extra clicks sites receive are no better and no worse than the clicks they already got. The randomized field experiment now shows a 39.8% drop in organic clicks on queries where an AI Overview appears, up from the 38% figure Search Engine Journal first reported in April. AI Overviews were triggered on roughly 41% of all queries tested. Across three separate engagement measures — return-to-results clicks, ten-second exits, and total time on site — the researchers found no statistically significant difference between clicks made with an AI Overview present and clicks made without one.

Does the study actually contradict Google's public explanation?

Yes. Google's own explanation for lost clicks doesn't hold up against the study's engagement data, which shows no measurable quality gap between the two groups.

According to Search Engine Journal's reporting on the paper, Google VP of Search Liz Reid has said publicly that AI Overviews reduce "bounce clicks" — visits so short and low-intent they add no value to the site receiving them. Bounce clicks is the industry's term for a click that ends almost immediately, typically measured by how fast a visitor returns to the search results page. Reid has not published data to support that claim. This experiment tested it directly by comparing the same three engagement signals — back-to-results returns, sub-10-second exits, and time on site — across both groups, and found about 4 in 10 same-tab clicks returned to results in both conditions, roughly 18% of visits in both groups ended within ten seconds, and time on site was statistically indistinguishable.

Why did the estimated click loss grow from 38% to 39.8%?

The researchers revised their methodology to include treatment-switch and query-type analysis, and the larger, more granular dataset pushed the estimate up by 1.8 points.

That gap matters more as a directional signal than as a precise number. A working paper that revises its own headline figure upward after adding analysis is telling buyers the initial estimate wasn't an overstatement — if anything, it undercounted. According to the authors, aggregate traffic losses are likely to grow further as AI Overviews expand to more queries over time, which means the 39.8% figure should be read as a floor, not a ceiling, for anyone forecasting organic traffic.

Which pages are actually losing traffic?

Informational content carries almost all of the measured loss, while navigational and transactional queries show little to no change in this dataset.

AI Overviews were triggered or intended to trigger on 53% of informational queries, compared with 15% of navigational queries and just 6% of transactional queries. That's the Query-Type Exposure Gap: the widening difference between how often AI Overviews appear on research-style searches versus purchase-ready ones. A site whose traffic leans on "what is," "how to," and comparison content sits in the 53% exposure zone; a site living on branded and transactional search sits closer to 6%. The position breakdown adds a second detail worth knowing — when a top-of-page AI Overview was removed, the first-ranked organic result nearly doubled its clicks, meaning the loss is concentrated at the top of the results page, not spread evenly down the list.

What should a publisher or SEO team actually do with this?

Stop accepting "AI Overviews only remove low-value visits" as a given, and start asking Google for the engagement data behind that claim.

Two things in the paper cut against the industry's default assumption. First, the consensus view — that falling click-through rate is offset by higher-quality remaining traffic — has no support here; quality was flat, not improved, in both directions of the experiment. Second, when the researchers rotated which group saw AI Overviews after the first two weeks, clicks per search moved in lockstep with the assignment, but zero-click rates mirrored each other in both groups — meaning zero-click growth isn't purely an AI Overview effect, which undercuts the industry habit of blaming AI Overviews for all zero-click search growth. This site has previously covered Google admitting its Search Console reporting for AI Search is inadequate; this paper is an independent, if unofficial, attempt to fill that reporting gap, and it's still just a working draft on SSRN that hasn't completed peer review.

Frequently asked questions

Is this a peer-reviewed study?

No. The paper by Saharsh Agarwal and Ananya Sen is a working draft posted on SSRN and has not completed peer review, so treat the figures as provisional.

Does a 39.8% click drop apply to every search?

No. It applies specifically to queries where an AI Overview appears; overall query-level impact is smaller because AI Overviews were triggered on about 41% of queries tested.

Did click quality actually improve when AI Overviews were shown?

No measurable improvement was found. Bounce-style returns, ten-second exits, and time on site were statistically indistinguishable between the AI Overview and no-AI Overview groups.

Sources: Search Engine Journal, "Google AI Overviews Study Finds Lost Clicks Weren't Lower Quality".