GEO does not replace SEO. It changes the rules.

A year of AI on the results page

Since 8 October 2025, AI Mode has been available in Switzerland, in the national languages as well as in English. AI Overviews, the summaries generated at the top of the results, were already live. France only followed in July 2026, after two years of deadlock over press neighbouring rights. Audiences in French-speaking Switzerland therefore have a year of usage ahead of their neighbour.

The scale is no longer experimental. At its I/O conference in May 2026, Google announced that AI Mode had passed one billion monthly users, and that AI Overviews reached 2.5 billion. Queries written in AI Mode are on average three times longer than classic searches. People no longer type “web agency Lausanne”. They describe their situation and ask a complete question.

To answer, Google breaks the question down into sub-queries, searches its index for each one, then assembles a summary with reference links. Google calls this mechanism query fan-out. It has a direct consequence: a page can be called upon for a sub-question nobody had written it for.

Fewer clicks, and different clicks

The most cited study on the impact of these summaries comes from the Pew Research Center. By analysing the real browsing of 900 American adults in March 2025, Pew found that users clicked on a classic result in 8% of visits when an AI summary appeared, against 15% when it did not. Links cited inside the summary were clicked in only 1% of cases.

Google disputes the methodology and states that clicks from pages with AI Overviews are of higher quality, with visitors spending more time on the site. Both findings can coexist. Simple informational queries, such as a definition or a general explanation, find their answer on the results page. The visits that remain arrive later in the journey. They are more qualified, and fewer.

For a B2B company or an institution, the stakes shift. An article that drew traffic by explaining a general concept loses part of its acquisition value. A page that documents a precise expertise, a concrete case or a reasoned position keeps its own. It can even gain value by becoming the cited source.

What Google says, in black and white

In May 2026, Google published a guide dedicated to optimising for its generative features. The message is clear. From Google Search’s point of view, optimising for generative search amounts to optimising the search experience, in other words doing SEO. The acronyms GEO (generative engine optimization) and AEO (answer engine optimization) are named, then tied to SEO rather than treated as separate disciplines.

The guide also dismisses several practices promoted by part of the market. Creating an llms.txt file or specific markup is not necessary to appear in AI Overviews or AI Mode. Nor is cutting content into small blocks (“chunking”), since Google’s systems can identify the relevant passage of a page that covers several topics. Rewriting texts “for AI” or multiplying keyword variants brings nothing specific. Structured data remains useful for rich results, but there is no schema.org markup dedicated to AI.

Conversely, Google insists on what it calls non-interchangeable content: an analysis, an experience or a piece of data that cannot be found elsewhere. On the technical side, the page must be indexed and eligible for a snippet. Semantic HTML is recommended, JavaScript rendering must follow best practice and duplicate content must be reduced.

One clarification is needed. This guide is about Google. ChatGPT, Perplexity or Copilot rely on their own crawlers and indexes, with criteria documented differently. OpenAI states, for example, that access for its OAI-SearchBot crawler determines whether a site appears in ChatGPT’s search results. A site that blocks this crawler out of caution removes itself from those answers.

From ranking to citation

If the basic rules do not change, the logic of performance does. On a classic results page, the question was where you ranked. In a generated answer, it becomes: is my page selected as a source, and for which part of the answer?

Three practical consequences follow. Covering a topic matters more than targeting an isolated query: a page that clearly answers the secondary questions of a field is more likely to be called upon by fan-out. Clear writing becomes a technical advantage, because an explicit definition, a sourced figure or an unambiguous step is easier to reuse correctly. Finally, the consistency of the entity carries more weight. The organisation’s name, its areas of expertise, its address and its public listings (Google Business Profile, professional directories, registers) must tell the same story.

Google also points out that its AI features can pick up what is said about a brand on blogs, videos or forums, while specifying that seeking artificial mentions hardly helps. Reputation is therefore built off the site as much as on it. And it is built slowly.

Measuring differently

Measurement remains the weak point of the moment. In Search Console, clicks and impressions from AI Overviews and AI Mode are included in overall search traffic, which makes their contribution hard to isolate. On the ChatGPT side, outbound links carry the utm_source=chatgpt.com parameter, which makes this traffic trackable in Matomo or Google Analytics.

Several signals therefore need to be combined: the trend of impressions against clicks, the share of brand and direct traffic, referral visits from AI assistants, and regular checks on a panel of questions representative of your business. None of these indicators is perfect on its own. Together, they give a more honest reading than simple rank tracking.

Frequently asked questions

Is GEO different from SEO?

For Google, no: its May 2026 guide ties optimisation for generative search to SEO. Other assistants (ChatGPT, Perplexity) use their own crawlers and criteria, which justifies also checking their crawlers’ access and the brand’s presence in their answers.

Should you create an llms.txt file?

Not for Google, which states that no specific file or markup is needed for AI Overviews and AI Mode. The file does no harm, but it replaces none of the fundamentals.

How do I know whether my site is cited by an AI?

By combining Search Console, tracking AI referrers in your analytics tool (for example utm_source=chatgpt.com) and regular tests on a panel of questions representative of your business.