Google's "Better Clicks" Claim for AI Overviews Just Failed a Controlled Test

Google's public defense of AI Overviews rests on three claims. The summaries make searchers more satisfied. They send higher-quality clicks to websites. And they help people discover more sites.

A new controlled experiment now tests all three claims directly. The results support none of them.

The research, covered by Digital Content Next on September 15, 2026, comes from Saharsh Agarwal and Ananya Sen. It is the first study on AI Overviews that can make a causal claim, because it does not just observe traffic falling. It removes the AI Overview for one randomly assigned group and measures what returns.

How the Experiment Worked

The researchers recruited 1,065 US desktop users and split them randomly into groups. One group used Google Search normally and saw an AI Overview whenever Google generated one. The other group installed a browser extension that stripped the AI Overview but left every traditional search result in place. Everyone searched the way they normally would.

This design matters. Almost everything we knew until now about AI Overviews came from referral data and behavioral logs. Publishers saw traffic fall after AI Overviews rolled out and inferred the cause. That inference was reasonable, but it was still correlation. I covered one of the stronger observational studies in a previous post about AI Overview citation clicks — one month of browsing logs from 900 US adults, where cited sources were clicked in only about 1% of visits. The authors of that study were careful to note it was not causal.

This new experiment closes that gap. When the AI Overview disappears and clicks come back, the argument that some other shift explains the traffic loss becomes very hard to sustain.

Claim One: Traffic. The Numbers Move Immediately

When the extension removed the AI Overview, outbound organic clicks rose from 0.37 to 0.62 per search. That is an increase of about 68%. Put the other way around, AI Overviews in their current form suppress roughly 40% of the clicks that would otherwise reach external sites.

Zero-click behavior moves with it. On pages with an AI Overview, 73% of searches ended without any click to an external site. Remove the Overview and that falls to 54% — still high, because Google keeps answers for many simple queries on the results page, but a 19-point gap attributable to the AI Overview alone.

We have tracked this zero-click drift for a while, from the early shift toward zero-click search marketing through the SparkToro clickstream series on Google search share and the question of whether Google Zero was real or a myth. Those analyses relied on panel data and inference. This experiment shows the mechanism directly: the AI Overview is not merely present while clicks fall. Removing it, clicks return.

Claim Two: Visit Quality. No Difference

Google's second claim is that the clicks AI Overviews do send are better ones — that a user who clicks after reading a summary arrives more qualified and more engaged.

The experiment found no measurable difference. Time spent on the destination site, bounce rate, and how often users came back to Google all looked the same whether the visit came from a page with an AI Overview or without one. The extra traffic that appears when AI Overviews are removed is, by every quality metric the researchers tracked, the same traffic.

This matters because "quality" is the consolation prize Google offers when publishers point at falling volumes. The pitch is: yes, you get fewer visits, but each visit is worth more. In this dataset, each visit is worth exactly the same. Fewer visits at unchanged quality is just fewer visits.

Claim Three: Satisfaction. Also Flat

The third claim is that searchers themselves prefer this. Google has argued since the AI Overviews launch in May 2024 that users report higher satisfaction with AI-enhanced results.

The researchers ran a post-study survey on exactly this point. Satisfaction with Google Search averaged 4.0 out of 5 in both groups. Ratings for information quality and for ease of finding information barely moved. Users did not like search more with AI Overviews. They did not like it less. They just clicked less.

So the summary box does not measurably help the searcher, and it demonstrably costs the publisher. Whoever the feature is optimized for, it is not the two parties Google names in its defense.

What About the Citations?

Google's usual reply is that AI Overviews cite their sources, and the citation links send traffic back. The experiment measured how much traffic that actually is. The answer: very little.

Take 100 clicks that leave a search page. On a normal results page, all 100 go to the organic listings. On a page with an AI Overview, fewer clicks leave in the first place. Of those that do, only about 7 come from the links inside the AI answer. The other 93 still come from the ordinary blue links below it. The citation box is a small pipe, attached to a feature that made the whole pool smaller.

Jason Kint, CEO of Digital Content Next, put the conclusion plainly in a thread on X: "Attribution without meaningful traffic or value is not a business model." The 7.1% figure is what that sentence looks like as a number. Being cited is real visibility — your name appears in the answer — but as a replacement for the clicks you lost, it is a rounding error. The feature cut outbound clicks by about 40%. The citations hand back roughly 7% of what remains.

Earlier observational data pointed the same way. The browsing-log study I discussed in the citation clicks article found that users clicked a cited source in only about 1% of visits. That study watched real behavior without changing anything. This one changed the page and measured the difference. Both land on the same conclusion: citations exist, but they do not replace traffic.

Why Causal Evidence Changes the Debate

Kint's thread makes one more point worth repeating: "Content became input. Traffic vanishes. Value flows to Google unless publishers define terms."

Until now, Google's available response to referral-data studies was that correlation is not causation — maybe queries changed, maybe user behavior shifted, maybe the sites that lost traffic deserved to. A randomized experiment forecloses that defense. The query mix and the users were the same in both groups. The only difference was the presence of the AI Overview, and the traffic followed it.

For publishers and SEOs, this study is the piece of evidence to cite when a stakeholder repeats Google's talking points.

• Satisfaction: no effect.
• Quality: no effect.
• Clicks: down 40%.
• Citations: 7.1% of what remains after all other links were removed.

The traffic impact is also not evenly distributed. It concentrates on a specific class of searches — informational queries, the backbone of most content marketing — and on AI Overviews placed at the very top of the page, where removing one restores 88% more clicks.

That distribution question, and what to actually do about it in a keyword and content strategy, is what I will work through in my next article.