AI Overviews cut clicks 39.8%: the first randomized study and what it changes for 2026 content strategy

#aioverviews #geo #aeo #seostrategy

AI Overviews cut clicks 39.8%: the first randomized study and what it changes for 2026 content strategy

Summary. A randomized field experiment with 1,065 desktop Chrome users found that Google's AI Overviews cut outbound organic clicks by 39.8% and raised zero-click searches by 34.5% when the summary appeared. On queries where an overview would show, hiding it lifted organic clicks per search from 0.37 to 0.62 and dropped the zero-click rate from 0.73 to 0.54. AI Overviews triggered on about 41% of observed queries, and more than 87% of them sat in the top slot above every organic result. The paper, by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University's Heinz College, was posted to SSRN on April 3, 2026, and last revised June 17, 2026. Its most awkward finding for Google: the clicks lost to AI Overviews were no lower in quality than the clicks that remained. With AI search advertising and content spend running into billions of dollars in 2026, and regulators in the European Union, the United Kingdom, and the United States weighing whether publishers should get an opt-out or compensation for AI summaries, the study gives that debate its first piece of randomized, causal evidence.

Until now, almost every number in the AI Overviews traffic debate came from observational data: compare traffic before and after the feature rolled out, or compare queries that trigger it against queries that do not. That approach cannot rule out that something else, a competing product or a Google algorithm change, moved the traffic. This study can, because it randomized who saw AI Overviews and who did not. For anyone still planning content as though the click loss were unproven, the argument just changed.

What the experiment actually did

The researchers built a Chrome extension on Webmunk, an open-source behavioral-research tool supported by the National Bureau of Economic Research, and used it to randomly assign real users to one of three conditions. A control group saw Google's standard results page. A treatment group had AI Overviews hidden by the extension whenever one would otherwise appear. A third group was redirected to Google's AI Mode conversational interface for every query. The two main groups, control and hide-overview, were each assigned with 36% probability; the AI Mode arm took the remaining 28%.

Data collection ran for two weeks per participant, from January 7 to February 10, 2026. Across 396 control users and 374 hide-overview users, the extension logged 68,089 unique searches, recording click behavior, scroll depth, and the full layout of each results page. Page-load times were statistically identical across groups, at 2.02 seconds for the hide-overview condition against 1.97 seconds for control, so speed does not explain the click gap. Because assignment was random, the only systematic difference between groups was whether an AI Overview was shown, which is what lets the authors call the effect causal rather than correlational. This is the distinction our own ranking versus AI Overview citation data analysis has been waiting for.

The headline numbers

For queries where an AI Overview was intended to appear, removing it raised outbound organic clicks from 0.37 to 0.62 per search, a jump of about two-thirds. The same removal cut the probability of a zero-click search, one that ends with no visit to any external site, from 0.73 to 0.54. Sponsored clicks did not move, and neither did clicks that stayed inside Google's own properties. That pattern matters: it says AI Overviews substitute specifically for organic website visits rather than reshuffling attention across the whole page.

Placement drove the damage. More than 87% of AI Overview appearances sat in the top position, above all organic listings, and the click effect concentrated there. Removing an overview from that top slot produced an 88% relative increase in outbound clicks, while overviews lower on the page had no measurable effect. Broken down by rank, position one gained the most, nearly doubling its clicks when the top overview was removed, with smaller but still significant gains at positions two and three.

Query type decided who felt it. Informational searches, 71% of the sample, carried almost the entire effect, with a statistically significant increase of 0.26 clicks per search when the overview was hidden. Navigational and transactional queries, 18% and 12% of the sample, showed no measurable change. AI Overviews also triggered far more often on informational queries, at 53%, against 15% of navigational and 6% of transactional ones. If your traffic is informational, this is your problem.

The finding that undercuts Google's defense

Google's standing answer to the traffic studies is that AI Overviews mostly absorb low-value clicks, so the visits publishers lose were not worth much. Google's Vice President of Search, Liz Reid, has described the remaining clicks as "higher-quality clicks." The experiment tested that claim from the other side, comparing three engagement measures on the sites users landed on: how often they hit the back button to return to search, how often a visit ended within ten seconds with no interaction, and total time on the page.

None of the three differed meaningfully between groups. About four in ten same-tab clicks returned to the results page in both conditions, roughly 18% of visits ended within ten seconds in both, and time on site was statistically indistinguishable. The paper states the result is "at odds with the view that AIOs primarily eliminate low-engagement website visits." In plain terms, the extra clicks a site gets when overviews are hidden look just as engaged as the clicks it already had. User satisfaction told the same story: an endline survey found the difference between groups on overall satisfaction, information quality, and ease of finding information to be, in the authors' words, "precisely estimated nulls." Removing AI Overviews cost users nothing they could report, and sent more traffic to publishers.

How this compares with the earlier evidence

The 39.8% figure sits toward the upper end of a range that observational studies have reported for two years, but it carries more weight because of how it was produced.

Study	Date	Method	Headline finding
Agarwal and Sen (SSRN)	April 2026	Randomized field experiment, 1,065 users	Organic clicks down 39.8%; zero-click up 34.5%
Ahrefs	April 2025	Observational, 300,000 keywords	Top result loses 34.5% of clicks with an overview
Ahrefs	February 2026	Observational follow-up	Reduction grew to 58% for top-ranking pages
Pew Research Center	July 2025	Observational browsing data	Users click traditional results about half as often
Seer Interactive	November 2025	42 client organizations, Jun 2024 to Sep 2025	Organic CTR down 61%; paid CTR down 68%

Google disputed the Pew figures as resting on "a flawed methodology and skewed queryset," and its Vice President of Search Nick Fox said in a May 2025 appearance that "from our point of view the web is thriving." The value of the randomized study is that it does not depend on comparing time periods or query sets at all. It compares the same users, in the same weeks, split only by whether a summary was shown. That is a different standard of proof, and it is why this paper is harder to wave away than the observational work that preceded it. For the strategy view built on that earlier data, see our ranking number one but getting no clicks breakdown.

What the paper does not claim

The authors are disciplined about their limits, and honest analysis has to be too. The AI Mode arm, which showed a steeper click drop to 0.36 per search against 0.53 for control and much lower satisfaction, is described as exploratory because it lost more participants to dropout, biasing its results. The position analysis captures a correlation with where Google chose to place an overview, not a causal effect of position itself, since Google's systems decide placement. The study measures clicks, not publisher revenue, and it does not estimate how the 41% trigger rate will change as Google expands the feature. The paper is a working draft on SSRN and has not completed peer review. None of that overturns the core result, but it should temper anyone tempted to treat a single number as the whole story.

What this changes for your 2026 content strategy

The strategic mistake now would be to read 39.8% as a reason to despair about clicks. It is better read as a map of where clicks are leaving and what to do instead.

First, stop treating all queries the same. The loss is concentrated in informational intent, where overviews trigger on 53% of searches, and is near zero on navigational and transactional intent. Content that exists purely to answer a simple informational question is the most exposed, and the least defensible, because an AI Overview reproduces it. Reallocate toward decision-stage, comparison, and transactional content where the overview rarely appears and the click still lands. This is the same logic behind our keyword selection framework for AI Overview click survival.

Second, compete for the citation, not just the link. Since overviews sit above organic results 87% of the time and absorb the visit, being the source the overview cites is now worth as much as ranking beneath it. That means structured, quotable, well-sourced content that answer engines lift cleanly, and off-site presence in the places they pull from, the approach we detail in earning third-party AI citations.

Third, change what you measure. Clicks alone now understate your visibility, because a brand can be seen and cited inside an overview without a click. Track impressions, AI-surface citations, and branded-search lift alongside clicks, using the Search Console AI performance report as the starting point. The study's own caution, that engaged clicks are still being lost, means the answer is not to abandon click optimization but to add citation and brand measurement beside it.

India-specific considerations

Indian publishers and D2C brands sit at the exposed end of this. AI Overviews and AI Mode have expanded across markets through 2026, and India's search-heavy small businesses lean on organic informational traffic for discovery in a way that makes a 39.8% cut on that intent a direct hit to top-of-funnel demand. Smaller sites tend to lose the most, which in the Indian context means independent publishers, regional-language content sites, and early-stage D2C brands feel it before large media groups do.

The practical response for an Indian marketing team is the same as the global one, applied with tighter budgets: move spend from thin informational content, which now feeds the overview rather than the site, toward comparison, pricing, and buyer-intent content that still earns the click, and build the branded demand that survives a zero-click results page. Measuring in rupees, the cost of continuing to produce answer-shaped content that an overview absorbs is real, and it is rising as coverage grows.

FAQ

What did the AI Overviews study find?

A randomized field experiment with 1,065 Chrome users found that Google's AI Overviews cut outbound organic clicks by 39.8% and raised zero-click searches by 34.5% when a summary appeared. On affected queries, hiding the overview lifted organic clicks per search from 0.37 to 0.62 and dropped the zero-click rate from 0.73 to 0.54.

Why is this study different from earlier ones?

Earlier studies from Ahrefs, Pew, and Seer used observational data, comparing traffic before and after AI Overviews or across query types, which cannot rule out other causes. This paper randomized which users saw overviews, isolating the causal effect. That randomized design is a stronger standard of proof than the correlational work that came before it.

Who conducted the research?

Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University's Heinz College. They posted the paper to the Social Science Research Network on April 3, 2026, and last revised it on June 17, 2026. It remains a working draft and has not yet completed peer review, a limit the authors state plainly.

Were the lost clicks lower quality?

No. The study compared back-button rate, ten-second bounce rate, and time on site between groups and found no meaningful difference. The paper says this is "at odds with the view that AIOs primarily eliminate low-engagement website visits," which contradicts Google's argument that the clicks publishers lose to overviews were low value.

Which queries lose the most traffic?

Informational queries, which were 71% of the sample and triggered an overview 53% of the time, carried almost the entire effect at 0.26 more clicks per search when the overview was hidden. Navigational and transactional queries, at 15% and 6% trigger rates, showed no measurable click change.

How should content strategy change in 2026?

Shift away from thin informational content that overviews reproduce, toward comparison, pricing, and buyer-intent content where overviews rarely appear and the click still lands. Compete to be the source an overview cites, and measure impressions, citations, and branded search alongside clicks rather than relying on click volume alone.

Does this affect Indian businesses?

Yes. AI Overviews expanded across markets through 2026, and India's search-dependent small publishers and D2C brands rely on informational organic traffic for discovery. A 39.8% cut on that intent hits top-of-funnel demand, and smaller sites tend to lose the most, so independent and regional Indian sites feel it before large media groups.

How eCorpIT can help

eCorpIT is a Gurugram-based, senior-led engineering and search organisation, founded in 2021 and CMMI Level 5 certified, that helps brands adapt content strategy to AI search. We audit where AI Overviews are taking your informational traffic, rebuild the exposed content toward buyer-intent and citation-ready formats, and set up the measurement that tracks AI-surface visibility, not just clicks, through our GEO and AEO content service. If your organic traffic is sliding while your rankings hold, talk to our search team about a citation-first plan.