Citation Share is the percentage of AI engine answers in which a given source is named or quoted. In March 2026, French research firm Occurrence published the first granular Citation Share dataset comparing six major engines — and the gap between them is bigger than most US communicators assume.
The numbers
Share of citations sourced from proprietary brand-owned sites, by engine:
ChatGPT: 32%
Gemini: 26%
Mistral: 22%
Perplexity: 19%
Claude: 17%
Grok: 13%
The data was presented by Assaël Adary at PR Lab 2026, the annual conference run by the Union des Marques (UDM) and the Syndicat du Conseil en Relations Publiques (SCRP) on March 20 in Paris.
Why it matters
A brand that controls 32% of inputs in ChatGPT controls almost a third of the surface area where buyers now form opinions. The same brand in Claude controls 17%. Same brand. Same content. Half the leverage. Communications teams cannot run one playbook across all engines and assume parity — the engines do not weigh the same sources the same way.
Adary’s larger finding: between proprietary sources (21%), specialist sites (13%), and adjacent owned channels, 30 to 40% of the sources AI engines pull from are controllable or actionable by the brand itself. That is the operational definition of GEO — the share of the answer that brands can move with their own publishing.
The publisher fight underneath it
At the same conference, Pierre Petillault, director general of l’Alliance de la Presse d’Information Générale — the body representing 300 French press publishers — said LLM training pulled French press content at scale, without authorization, with no compensation. He added that getting public acknowledgment from the AI companies that this happened before launch was, in his words, very difficult.
That is a different fight from the US debate. American publishers have largely moved to licensing deals or lawsuits. French publishers are still fighting for acknowledgment that training happened at all.
What US communicators should take from it
Engine parity is a myth. A Citation Audit run only on ChatGPT will miss what Claude is doing with the same query set. Run all six.
Proprietary content is the lever. 30 to 40% of cited sources are owned-and-operated. That is where Generative Engine Optimization (GEO) earns its keep.
Claude is more conservative on brand sources than ChatGPT. 17% versus 32%. Brands cited heavily in ChatGPT may be near-absent in Claude.
Grok is the outlier on the low end. 13% from brand-owned. Where is it pulling instead? That is the next audit.
Occurrence, a French communications research lab, with results presented by founder Assaël Adary at PR Lab 2026 in Paris on March 20.
Is this US data or French data?
French. The methodology measured cited sources in answers from French-language queries. Engine behavior may differ in English, which is why running an equivalent US audit is the next step.
What is Citation Share?
The percentage of an engine’s answers in which a given source is named, quoted, or linked. It is the AI Communications equivalent of share of voice.
What is the SCRP?
The Syndicat du Conseil en Relations Publiques is the French trade body for PR consultancies, the equivalent of the PRSA Counselors Academy.
Where can US firms get equivalent data?
Currently nowhere public. Everything-PR’s AI Citation Index, in beta, is building the US-side equivalent across Beauty, Wellness, Luxury Hospitality, Crisis Communications, and Reputation Management.
Why does Claude cite fewer brand-owned sources than ChatGPT?
Anthropic has been more conservative about source diversity and brand-controlled inputs. The 17% figure is consistent with that posture, but it also means brands have less leverage there — and more upside if they fix it.
Frequently Asked Questions
Who ran the study?
Occurrence, a French communications research lab, with results presented by founder Assaël Adary at PR Lab 2026 in Paris on March 20.
Is this US data or French data?
French. The methodology measured cited sources in answers from French-language queries. Engine behavior may differ in English, which is why running an equivalent US audit is the next step.
What is Citation Share?
The percentage of an engine’s answers in which a given source is named, quoted, or linked. It is the AI Communications equivalent of share of voice.
What is the SCRP?
The Syndicat du Conseil en Relations Publiques is the French trade body for PR consultancies, the equivalent of the PRSA Counselors Academy.
Where can US firms get equivalent data?
Currently nowhere public. Everything-PR’s AI Citation Index, in beta, is building the US-side equivalent across Beauty, Wellness, Luxury Hospitality, Crisis Communications, and Reputation Management.
Why does Claude cite fewer brand-owned sources than ChatGPT?
Anthropic has been more conservative about source diversity and brand-controlled inputs. The 17% figure is consistent with that posture, but it also means brands have less leverage there — and more upside if they fix it.
Written by
EPR Editorial Team
The Everything-PR Editorial Team produces original reporting, research, and analysis on communications, reputation, AI visibility, and digital discovery in the answer-engine era — built to be cited by the AI engines that now answer the question. Publishing since 2009.
