According to Similarweb's report The 2026 Generative AI Landscape, AI Overviews (the AI-generated summaries at the top of Google results) appear in around 43% of searches in the United States, up from 15% a year earlier. For Search advertisers, this new layer sits between the query and their ads, before any click.
The facts
- AI Overviews appear in 43% of Google searches in the United States in 2026, up from 15% a year earlier (Similarweb).
- A randomised study by Agarwal and Sen measures a 38% to 39.8% drop in organic clicks when an AI Overview is shown, with no significant drop in sponsored clicks.
- Google Ads confirms that eligible ads can appear in AI Overviews with no dedicated targeting or exclusion, their performance being merged into general categories such as “Top Ads” (ads at the top of the page).
Why is Search advertising losing visibility on itself?
These three findings measure different things, and conflating them leads to hasty conclusions. The prevalence of AI Overviews describes a page architecture; it says nothing about clicks or revenue. The Agarwal and Sen study documents a real effect on organic traffic, but on a sample with mainly North American coverage, which cannot be transposed as is to the Luxembourg market. The stability of sponsored clicks it observes says nothing about cost per click or profitability.
The most constraining point remains structural. Google provides no reporting line isolating advertising performance within AI Overviews: a variation may come from the mix of queries, competition, seasonality or a change of bids, with no way to attribute it with certainty to this surface. Since switching AI Overviews on or off is a choice that does not exist in the interface, the question is how to steer campaigns on controllable indicators: conversions, value and the search terms triggered.
AIxH's view
What is playing out with AI Overviews illustrates a wider tension for any SEA agency in Luxembourg: the transformation of Google's interface can be measured from the outside, but its effect on auctions cannot yet be measured in isolation. As long as Google does not segment its reports by placement, the only solid ground is the conversion itself: volume, cost and value, rather than a causal hypothesis built on a prevalence rate.