Why does ChatGPT cite public administration rather than businesses?
Published on
Because, on questions relating to a rule, taxation, or health, ChatGPT's search retrieves many public websites, and it then cites them more readily than other pages. According to our corpora, public websites represent 15% to 71% of its citations, and a public page retrieved by its search is cited 1.4 to 3 times more often than any other. Claude and Gemini cite them much less, making up less than 10% of their citations. A company cannot replace public administration on rules; but it can be the source for what public administration does not say: concrete application, specific cases, costs, and options.
For what type of question? It all depends on the question. On tax, regulatory, or health questions, public websites dominate in ChatGPT. On expertise, craftsmanship, or product choice, they are absent, and the floor goes to specialists.
What we measure in ChatGPT
We classified as public websites the domains of French, American, and European administrations, including public health databases. In ChatGPT, they accounted for 71% of citations on our HVAC questions in French, 29% across ten business sectors in both French and English, 23% on professional software troubleshooting, and 15% on artificial intelligence in English. On troubleshooting questions that did not name any brand, a manual classification even attributes 45% of ChatGPT's citations to public administration, such as Chorus Pro or impots.gouv.fr.
Two mechanisms combine. First, ChatGPT's search retrieves many public pages: from 9% to 45% of retrieved pages depending on the corpus. Second, once retrieved, these pages are chosen more often: they are cited in 13% to 21% of cases, compared to 7% to 10% for other pages, which is 1.4 to 3 times more. Among the most cited public sites in our sectors in French were impots.gouv.fr, American public health databases, INAO, Légifrance, and service-public.fr.
It all depends on the subject of the question
Across the ten sectors queried in French, the share of public websites in ChatGPT's citations ranged from 71% on beauty, where it cited public medical databases, and 59% on heritage, taxation, and inheritance, to 42% on agriculture, 38% on education, 30% on health, 17% on old buildings, 16% on professional software, and 7% on beekeeping. It was zero on lutherie, fishkeeping, and visibility in AI.
The boundary is clear: as soon as a question touches on a rule, a tax, a regulation, or health, ChatGPT relies on the official source. When it concerns expertise or practice, there is no official source to cite, and specialists take over.
Claude and Gemini behave differently
For Claude, public sites represented 0% to 11% of the retrieved pages and 0% to 8% of citations depending on the corpus; for Gemini, 1% to 5% of sources. On troubleshooting questions without brands, where ChatGPT attributed 45% of its citations to public administration, manual classification shows 19% for Claude and 18% for Gemini, which cite software publishers more.
The preference for public administration is therefore primarily a characteristic of ChatGPT and the search it uses. A company absent from ChatGPT's answers on a regulatory topic may be present in Claude or Gemini on the same question.
What Google says about trust
Google explains, in its helpful and reliable content guidelines, that among expertise, experience, authority, and trustworthiness, trust is what matters most. It specifies that its systems give even more weight to content that inspires trust on topics that can affect people's health, financial stability, or safety, which it calls "Your Money or Your Life".
These recommendations apply to Google Search, not to ChatGPT. Yet they describe the same logic as our measurements: on a topic where a mistake is costly, the official source wins. It is precisely on these topics that ChatGPT cites public administration the most.
What this changes for a business
Do not try to replace the official source on the rule itself: on this ground, ChatGPT prefers it. Occupy what the administration does not cover: how the rule applies to a concrete case, what it costs, common mistakes, the choice between several options, execution. A page that cites the official source and explains it for a specific case complements the administration instead of competing with it.
Also look at the type of questions from your clients. If they are mostly about rules, expect to share the space with the administration on ChatGPT, and look after your presence on Claude and Gemini. If they are about know-how or a choice, the space is free for specialists, and this is where your pages have the highest chance of being cited.
What we do not know
Our ranking of public sites is based on domain extensions (.gouv.fr, .gov, europa.eu): it includes American public medical databases and ignores public organizations without this extension. The shares by sector are based on few citations each. Our measurements cover the GPT-5 API, not the ChatGPT application.
Sources
- Google Search Central — "Creating helpful, reliable, people-first content", .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see "How we measured".