Causal Study: Hiding AI Overviews Raises Clicks 38 Percent

A randomised field experiment with 1,065 US desktop Chrome users found that hiding Google AI Overviews raised outbound organic clicks from 0.38 to 0.61 per search, a 38 percent suppression on triggered queries, and cut zero-click searches from 72 percent to 54 percent, with no measurable change in user satisfaction and over 95 percent of treated users noticing nothing. The finding anchors a November 2026 update to a GEO statistics compilation, citing Agarwal and Sen, 2026.

Why this number is different

Most AI search statistics are observational: they compare pages with and without AI Overviews, but those pages differ in many ways besides the overview. A randomised field experiment removes that problem by assigning users to see or not see overviews, so the difference in clicks can be read as caused by the overview itself. That is why the compilation calls 38 percent the single most useful GEO number in 2026: it is causal, not correlational.

The satisfaction result is equally important. Over 95 percent of treated users noticed nothing, and satisfaction did not measurably change, which undermines the argument that users demand AI summaries even at the cost of publisher traffic. If users are equally satisfied without the overview, the case for showing it on every query weakens considerably.

What the wider statistics update adds

The same November 2026 update adds a retrieval-versus-page-features split: retrieval accounts for 32 percent of within-site citation variation against 2.8 percent for page features, citing Chen and Koudas, arXiv 2610.15033. Pages updated in the last 12 months had 88 percent higher citation odds, according to TJ Digital, and product pages were the most-cited page type at 28 percent.

The supply-side finding is stark: only 2.9 percent of AI citations point at the cited brand's own domain, while 75.2 percent point at third-party pages, according to Ranqo Labs analysis of 149,912 citations. Together, these numbers describe a system where being retrievable matters far more than page polish, and where third parties mediate most of the citation value.

What SEOs should do with a causal number

First, use the 38 percent figure honestly: it applies to triggered queries in this experiment, not to all searches, and it measures outbound organic clicks per search, not revenue. Quoting it as a universal traffic loss overstates the study, and overstated claims undermine the credibility publishers need in policy debates.

Second, act on the retrieval finding. If retrieval explains 32 percent of citation variation and page features only 2.8 percent, the highest-return work is making pages retrievable: clear topical focus, fresh updates, and content that matches the queries assistants actually fan out to. Page polish still matters, but it is not where the citation battle is won.

From causal evidence to publisher strategy

Causal evidence changes the negotiation between publishers and platforms because it replaces argument with measurement. A publisher armed with a randomised 38 percent suppression figure can make a concrete case to regulators, to advertisers, and to its own board, in a way that anecdotes about falling traffic never could. Expect this study to be cited in policy debates, licensing negotiations, and earnings calls well beyond the SEO community.

For individual site owners, the study also sets a benchmark for their own testing. If your own opt-out test shows suppression far above or below 38 percent, the difference is informative: it may reflect query type, audience, or content format. Running that test, with a control group and a pre-registered metric, is now the standard of evidence, and sites that can produce it will be taken more seriously than sites that can only report that traffic feels lower.

Sources

Sources: GeoAura GEO Statistics, November 2026 update, citing Agarwal and Sen, 2026.