How an AI picks the sites it cites

The assistant doesn't search your question as is. It rewrites it into one or more queries, often more precise, sometimes in English even when the question was in French. It then queries a search index, its own or a partner's.

The direct consequence: if your page is not indexed, or ranks poorly on those queries, it does not even enter the race. GEO starts with basic SEO in good shape.

From the results, the assistant opens a handful and reads their content. A page that is slow, blocked to crawlers or whose text only appears after running a lot of JavaScript risks being read halfway, or not at all.

• The important text is in the HTML, not only in an image or a video.
• AI crawlers are not blocked by robots.txt or by a firewall.
• The page loads fast, with no pop-up hiding the content.

The model then looks, in those pages, for the passages that answer the question most directly. A paragraph that opens with the answer, with a number, a place, a price or a date, is easier to reuse than text that circles around the topic.

This is where writing style counts. A heading phrased as a question, followed by a two-sentence answer, then the details: that format helps a reader in a hurry, human or machine.

The sources shown are generally the pages whose passages fed the answer. Being cited means you supplied a useful piece. Being named without a link often comes from elsewhere: what other sites say about you, directories, reviews, press articles.

Three priorities, in this order: be indexed and readable, write clear answers to the real questions your customers ask, and get talked about on sites that AI consults. Nothing magic, but a different order of priorities than classic SEO.