How does ChatGPT choose the sources it cites in France?

When a generative AI answers a question, it draws on very specific sources — but which ones, and following what logic? For the first time, our proprietary Search Ai study lifts the veil on the real hierarchy of domains cited by ChatGPT in France. Conducted across more than 100,000 sources and 20+ business sectors, it challenges several assumptions about SEO and opens the way to a new discipline: GEO, or Generative Engine Optimization.

Methodology: a database of more than 100,000 cited sources, 100% French

To analyze what generative AIs actually cite, we built a first-of-its-kind database for the French market. The scope rests on four pillars: more than 100,000 sources analyzed, 20+ business sectors covered, 100% of queries formulated in French, and ChatGPT as the main conversational engine — the most widely used today.

Data processing follows a rigorous four-step method. First, extracting the citations present in the generated responses. Next, aggregating by root domain, so that URLs from the same site are grouped under a single entity. Then, calculating each domain's share of citations. Finally, a complementary analysis of «query fan-out» and AI bot access, to understand not only which sources are cited, but also how AIs build their queries upstream.

The real ranking: Reddit and public services crush the competition

The first finding, and not the least significant, is that the hierarchy of sources cited by ChatGPT has little to do with a traditional SEO ranking based on domain authority. Reddit comes out well ahead, followed by public .gouv.fr sites. Together, the two account for 77.5% of the Top 10 citations.

"TOP 10 cited sources in France - ChatGPT", Source Search Ai

Behind this dominant duo come the major platforms and the press (Google, Le Monde, Wikipedia, Le Parisien), then review sites and marketplaces (Trustpilot, Quechoisir, YouTube, Leboncoin). Two distinct logics emerge: on one side, Reddit's community-driven, conversational content, rich in authentic first-hand accounts; on the other, the perceived institutional reliability of public sources.

Beware the misleading effect of root domains

These three major domains actually conceal a multitude of subdomains grouped under a single ranking line. For gouv.fr, ChatGPT draws on service-public.gouv.fr, economie.gouv.fr, ecologie.gouv.fr, entreprendre.service-public.gouv.fr, and culture.gouv.fr just as much as on legifrance.gouv.fr: each ministry or administration is a citable source in its own right. The same is true for wikipedia.org, which covers both fr.wikipedia.org and en.wikipedia.org. Google is the most fragmented case: support.google.com, maps.google.com, scholar.google.com, developers.google.com, cloud.google.com, and blog.google are all distinct subdomains, well beyond the google.com search engine alone. In other words, the “share” of a root domain masks a diversity of content that needs to be analyzed carefully.

The number that redefines strategy: 7.2%

Here is probably the most counter-intuitive result of the study: the entire Top 10 domains represent only 7.2% of total citations. The remaining 92.8% is spread across a long tail made up of hundreds of different domains.

"All sources ChatGPT France", Source Search Ai

This figure radically changes the way we think about a visibility strategy. Being cited by a generative AI is not just about appearing among a handful of high-authority media outlets. Betting solely on a few premium domains therefore means ignoring more than nine citations out of ten.

For a brand, the challenge becomes existing across a broad, relevant network of sources, rather than concentrating all efforts on a limited number of high-profile sites. This is a complete paradigm shift compared with traditional link-building approaches.

The hierarchy varies significantly by sector

The Top 10 presented here is an average calculated across 20+ sectors. Within a given sector, the reality can be very different. Our sector-by-sector analyses show marked variations:

• Health: reference medical sources (such as Doctissimo and specialized medical sites) rank ahead of general-interest media.

• Finance: institutional portals and comparison sites dominate, far ahead of news media.

• High-tech: review and comparison media (product tests) rank ahead of Wikipedia.

The consequence is direct: there is no universal recipe. Identifying the sources actually cited in your own sector is a prerequisite for any GEO strategy.

Query fan-out: decoding what ChatGPT is really looking for

Beyond the sources cited, Search Ai analyzes «query fan-out»: when an AI receives a question, it breaks it down into several sub-queries to retrieve information. By observing the words that recur most often in these sub-queries, we gain a better understanding of how AI search actually works.

"TOP words in the Query fan-out - GhatGPT", Source Search Ai

Three major findings emerge from this data. First, a massive local anchoring: the word “france” appears in 41.8% of sub-queries, even on seemingly generic topics. Second, a demand for freshness: “2026” appears in 10.2% of sub-queries, proof that AIs favor recent, dated content. Finally, a clear commercial intent: the terms advice, reviews, price, advantages, best, and choose are among the most frequent, a sign that AIs are actively seeking to compare options and support decision-making.

For a content publisher, these signals translate into concrete recommendations: systematically contextualize content for the French market, adopt a comparative, decision-oriented angle rather than a purely descriptive one, and regularly update pages with explicit time markers.

What these results mean for your GEO strategy

In summary, the Search Ai study outlines three priorities for being cited by generative AIs. First, diversify your presence across a large number of relevant sources, rather than targeting only the dominant domains, since the Top 10 accounts for just 7.2%. Second, align your content with the phrasing actually used by AIs: a comparative tone, highlighting reviews, prices, and advantages, and keeping information fresh. Third, think sector by sector, since the hierarchy of cited sources differs significantly from one line of business to another.

GEO is no longer a theoretical hypothesis: it is a measurable discipline, based on concrete citation data. Brands that incorporate it into their content strategy now are gaining a head start over those that continue to think solely in terms of traditional SEO.

Take action with an analysis dedicated to your sector

Want to know exactly which sources ChatGPT cites in your sector, which words dominate your sub-queries, and how to steer your content production accordingly? Request a demo of our Search Ai tool: we'll walk you through a personalized analysis of your visibility in generative AI.