Google AI Overviews: what they do to your traffic

AI Overviews, which Google calls Aperçus IA in French, are the generated summaries displayed above the results for certain searches. One question keeps coming up among business leaders: how much traffic do they take from websites, and should anything change?

:::figure{src=“/blog/ai-overviews-trafic/apercu-ia-definition.webp” alt=“Google.fr results page on a phone for “qu’est-ce que le référencement naturel” (what is SEO): an AI Overview at the top of the screen gives the full definition, four related questions follow, and the first website, bpifrance creation, only appears at the bottom of the screen.” cadre=“telephone”} On google.fr, the definition of SEO comes from the AI Overview, and the first website only appears at the bottom of the screen, after four related questions. Captured on a phone on 13 October 2026. :::

Here is what independent data measures, then how to check it on your own site. This article extends our guide to what an AI SEO agency should deliver, which sets the framework for this section.

How many clicks do AI Overviews cost?

The figures that follow come from the Pew Research Center study published in July 2025. It followed 900 American adults, members of a panel who had agreed to share their browsing, during March 2025: 68,879 Google searches, of which 12,593 displayed an AI summary. Links placed in the summary itself were clicked in only 1% of visits. The other measurements show where the effect lies:

The last comparison is the most useful for a business. About one search in five produced a summary, but the effect is not uniform: it hits questions, much less short searches, such as a brand or product name.

Two cautions when reading it. The study covers American users in March 2025; the rollout, the language and habits differ in Europe, and Google keeps changing the display. Second, Google disputes this type of conclusion: in its documentation on AI features in Search, it states that clicks from pages with AI Overviews are “higher quality”, in the sense that visitors are more willing to spend time on the site. That is the publisher’s own claim, with no published data to support it on that page. Both can be true at once: fewer clicks, but more engaged clicks.

Can you see AI Overviews traffic in Search Console?

Not directly. According to the same Google documentation, appearances in AI Overviews and in AI Mode are counted in the Search Console Performance report, under the “Web” search type, with the rest of the traffic.

In practice, an appearance in an AI summary counts as an impression, and a click from the summary counts as a click. Nothing distinguishes them from a traditional result. Hence confusing curves: impressions rising, because the page is cited more often, while the click-through rate falls, because the answer is enough for the reader.

How do you know whether a drop comes from AI Overviews?

Since Search Console separates nothing, you work by deduction, in four steps:

1. 01Isolate the queries phrased as questions.

   In the Performance report, add a filter on queries, in “Custom (regex)” mode. An expression such as ^(how|why|what|which|is|should|can|where|when)\s catches most questions in English. The Search Console help on filtering specifies that the syntax is RE2 and that matching is case-insensitive by default. The same filter set to “Doesn’t match regex” gives the group of the other queries.

2. 02Compare two periods, at equal position.

   The date filter offers a “Compare” mode, which adds a “Difference” column: take the same months year on year, to neutralise seasonality. If the click-through rate of questions falls while their average position holds, the most probable cause is a display that answers in place of the site, AI Overviews or another answer block.

3. 03Drill down to page level.

   With the same filter, the “Pages” tab shows who is losing. An overall drop often hides two or three informational articles that lose a lot, and service pages that do not move.

4. 04Check by hand.

   Type the queries that lose the most, from the target country and in a private window, and note which ones actually display a summary.

This diagnosis remains an inference. A click-through rate also falls when neighbouring results change or a competitor publishes something better. You need the clues together to conclude: questions that lose clicks, a position that holds and a summary actually displayed on those queries.

Which pages are exposed?

From these data, the pages of a business website can be sorted into three groups.

Definition pages and simple-answer pages are the most exposed. “What is a CRM”, “difference between SEO and PPC”: the AI summary answers in four lines, and some readers stop there. These pages were often written to capture traffic, rarely to convert.

Experience and method pages hold up better. A case study with figures, a comparison made on your own projects, a detailed procedure: the summary can cite the page, but it does not replace reading when the reader has to act.

Commercial-intent pages are the least affected. An offer, pricing, contact or case study page answers someone who already wants to buy or talk to someone. The short searches of one or two words that often lead there trigger far fewer summaries according to the Pew study.

What should you do when AI Overviews take clicks?

There is no magic setting. Google says so in the same documentation:

The SEO basics apply. To appear as a link in a summary, a page must be indexed and eligible to be shown in Search with a snippet. Here is what we recommend, in order.

Measure before reacting. Do the split between questions and other queries described above. Many companies worry about a drop that comes from elsewhere: a redesign, a deindexed page, a seasonal change.

Accept being the cited source. A well-structured page, with a clear answer at the top and sourced facts, is more likely to be picked up in the summary with a link. It is real visibility even without a click, and sometimes the only contact with a prospect who will come back later through a brand search. We detail this method in our guide to getting cited by AI search engines.

Give a reason to click. A tool, a template to download, a calculator, original figures that the summary cannot reproduce in full. The page that gives more than the short answer keeps its readers.

Shift effort towards the pages that convert. If your definition articles lose clicks but your offer pages hold, that is not necessarily bad news. It is better to strengthen the service pages and case studies than to rewrite ten definitions.

Where to start

Start with your own figures: the split described above tells you whether AI Overviews are weighing on your traffic or whether the drop comes from elsewhere. To then track your presence in assistants’ answers, and not only at Google, apply a protocol for measuring your visibility in AI.

If you prefer an outside view, our SEO audit reads your traffic figures over sixteen months and delivers a prioritised action plan within two weeks. To then produce, every month, the pages that answer those questions, our SEO and GEO content engine writes content built to be cited, and nothing goes out without your review.

Frequently asked questions

Sources

1. The Pew Research Center study published in July 2025 (new tab)pewresearch.org
2. Its documentation on AI features in Search (new tab)developers.google.com
3. The Search Console help on filtering (new tab)support.google.com