AI Overviews Now Appear in 43% of US Searches

A year ago an AI Overview was something that happened to some of your queries. It now happens to close to half of them. Similarweb's 2026 Generative AI Landscape report states that more than 40% of US searches trigger an AI Overview. TechCrunch, reporting on the same analysis on 27 July 2026, puts the number at 43%, up from 15% twelve months earlier. Roughly a tripling in a year, and the overview has quietly become the doorway to a second Google surface that behaves nothing like the first.

Key takeaways

The figure is the share of searches on which Google generates an AI Overview at all, measured across US queries. It says nothing about whether the user read it, and nothing about who got cited in it.

It is also worth being clear about provenance. This is panel-based measurement from a traffic analytics vendor, not a disclosure from Google, and competing estimates for the same claim land anywhere from the low twenties to the high forties depending on which queries a study samples and how it detects an overview. Navigational searches, branded lookups, and image queries mostly do not produce one; informational and commercial questions overwhelmingly do. A study weighted toward the latter reports a higher share, and both numbers can be honest.

What survives the disagreement is the direction and the pace. Every credible measurement of the last year points the same way, and a jump from 15% to 43% is far too large to be an artefact of methodology. The safe reading: for the kinds of questions your buyers actually ask, an AI answer is now the default presentation, not an occasional interruption.

The more consequential number in the same data is AI Mode's growth. Visits went from 126 million in June 2025 to 279 million in May 2026, more than doubling in under a year, and TechCrunch describes the mechanism plainly: Google drops users into an AI Overview, and from there they continue researching in the more conversational AI Mode.

That is a funnel, and it has a measurement consequence. The overview is a one-shot answer to the query as typed. AI Mode is a conversation, where each turn arrives with the earlier turns still in context. Those are different retrieval problems, and a brand can win one and lose the other.

Treating "Google" as one tracked engine averages these together and hides which one is failing. Split them.

Buried in the same report is a small number with an outsized implication: average Google search length is up 5.4% since AI Mode launched. People are typing sentences where they used to type fragments, because sentences now work.

Most tracked prompt sets have not caught up. They were assembled from keyword research, so they are full of short category phrases like "best CRM" rather than the request a buyer actually makes. That gap matters more than it looks, because the qualifiers people add are precisely the ones that decide a shortlist. Recent research on multi-turn AI search found the final prompt in a conversation carries a median of only about 36% of what the user said across the session, with a stated budget, location, or rejected option living in the history rather than in the prompt being answered half the time.

The fix is cheap. Go through your prompt set and count how many entries read like a keyword and how many read like something a person would type into a chat box. Then write the missing siblings: the same question with a price ceiling, a team size, a use case, an option ruled out.

Pew Research Center gave the click question its cleanest answer, and it did so back when the surface was much smaller. Analysing the browsing of 900 US adults in March 2025, Pew found users clicked a search result on 8% of pages carrying an AI summary, against 15% of pages without one. They were also likelier to abandon the session outright: 26% of pages with a summary ended the browsing session, versus 16% without.

Halve the click rate, then apply it to nearly three times as many searches, and you have the mechanism behind every publisher traffic chart from the past year. It also explains why AI referral traffic is such a poor scoreboard. When referrals fall, you cannot tell whether you lost standing in AI answers or simply held it on a surface that clicks less.

Which is why the counterweight in the same report deserves attention: Similarweb measures a 2.5x higher chance of a site visit after a brand is mentioned in an AI answer. The mention is doing work even where the link does not. So measure the mention. Track whether you are named, whether you are cited, and who is named alongside you, and treat any referral data you can get as supporting evidence rather than the number you report.

Google added a Search Console control to remove a site from AI Overviews and AI Mode in June 2026, and the case against using it was already strong: opting out does not stop Google answering the query, it only removes you as a source, so the answer gets written from a competitor's page instead.

At 43% of US searches, that concession is roughly three times the size it was when the surface stood at 15%. The argument for opting out has always rested on pageview-dependent revenue and on negotiating leverage with Google. Neither of those got stronger this year. The cost of the trade did.

Our own 42-day citation study is the useful complement to this data, because it describes behaviour rather than reach. Google AI Mode grounded every prompt on every day, cited about 24 domains daily, and concentrated roughly 56% of citations on a small core of sources that appeared almost every day. ChatGPT grounded selectively, mainly on commercial intent, and rotated even its leading sources, so its stable core captured only about 23% of citations.

That difference sets strategy. Google's AI surfaces reward becoming part of a durable core: hard to break into, hard to fall out of once you are there. ChatGPT's faster rotation gives newcomers more openings and offers less protection. Same brand, two games, and the reach numbers above mean the Google game is now the one played on nearly half of US searches.

One more sign of where these surfaces are heading, from the same report: Similarweb found advertising present in 26% of US desktop ChatGPT chats in June 2026. Answer surfaces are becoming inventory. The organic citation slot is not going away, but it is going to have neighbours.

Elmo is an open-source, self-hosted AI visibility platform that runs your prompt sets across ChatGPT, Claude, Gemini, Perplexity, and Google's AI surfaces, recording every mention, citation, and competitor named alongside you. Because Google AI Overviews and AI Mode are tracked as distinct engines and the data is yours to query, the split described above is a filter rather than a project.

New to this? Start with answer engine optimization, then how to track your brand in AI search. For the vocabulary, see the AI search glossary.