In France, ChatGPT crushes the competition: it accounts for 76% of the time spent on AI services, far ahead of Claude (5%), Gemini (4%) and Perplexity (3%). That is what the first AI services audience barometer from Arcom, the French media regulator, shows (published in June 2026, in French). For a brand, the takeaway is simple: ChatGPT is the priority, but a serious visibility measurement must cover every AI engine, each at its real weight.
Until now, marketers steered their AI strategy with global figures or vendor estimates. This barometer changes that. It is an independent audience measurement, published by the regulator, covering the French population. Here is what it says, what it does not say, and what you can do with it.
What is the Arcom AI barometer?
It is an audience measurement of AI websites and apps in France, published by Arcom using Médiamétrie//NetRatings data. It counts who uses these services and for how long. It says nothing about what people actually ask.
Arcom, France's audiovisual and digital communication regulator, published this first edition on June 16, 2026. The study relies on Médiamétrie//NetRatings internet audience measurement. The main results cover April 2026, with historical data going back to 2023.
Arcom's stated goal is not marketing. According to the regulator's announcement, the study sheds light on audience protection issues (especially for minors) and prepares its future responsibilities under the EU AI Act. But its figures are valuable for any company wondering where its customers ask their questions.
How many French people use AI in 2026?
Nearly 33 million French people used an AI service in April 2026, or 57% of the population. That is seven times more than in April 2023. And each user spends more and more time on it.
Key figures published by Arcom (source, in French):
- Nearly 33 million users of an AI service in April 2026, or 57% of the French population.
- 4.3 million in April 2023: usage grew sevenfold in three years.
- Average time per visitor: 19 minutes per month in 2023, 52 minutes in 2024, 1 h 26 min in 2025.
- More than three quarters of 12-17 year-olds used an AI service in April 2026.
- Time spent on AI services still represents only 1% of total time spent on the internet.
One point stands out for marketers. In its presentation video, Arcom states that French people now spend more time on AI services than on search engines. Information seeking is shifting. That is the whole point of GEO compared to SEO.
Which AI engines do French people really use?
ChatGPT, by far. It was used by 27 million French people in April 2026 and captures 76% of the total time spent on AI services. Claude, Gemini and Perplexity share 12% of that time between them.
| AI service | Share of total time spent on AI services (April 2026) |
|---|---|
| ChatGPT | 76% |
| Claude | 5% |
| Gemini | 4% |
| Perplexity | 3% |
Source: Arcom, AI services audience barometer, 1st edition (June 2026), Médiamétrie//NetRatings data.
Three lessons for a brand:
- ChatGPT is the main playing field. Three quarters of the time spent on AI in France happens there. If your brand is missing from ChatGPT's answers, it is missing from most conversations.
- Audience share and time share tell different stories. A service can reach many people while spending little time with each, or the opposite. Time spent reflects intensity of use, not just awareness.
- Claude is ahead of Gemini in time spent. This is counter-intuitive: Gemini benefits from Google's ecosystem. But the barometer measures time spent on the services themselves, not AI built into other products (see the limitations below).
What are the limitations of this barometer?
The barometer measures time spent on AI websites and apps. It does not count the number of questions asked, professional use through APIs, or AI built into other products. It is a reliable snapshot, but a partial one.
Before reallocating a budget, keep four limitations in mind:
- Time spent is not the number of questions. A long writing session on ChatGPT weighs more than a quick question on Perplexity. Yet the quick question ("what is the best software for…") is often the one that drives a purchase. Time share overweights long sessions.
- Built-in AI is not counted. According to Arcom's presentation, the scope covers dedicated AI websites and apps, not features natively built into devices or other services. Google AI Overviews, the assistant built into Windows or AI inside messaging apps are therefore outside this measurement. Yet they reach a huge audience.
- API and integration usage is invisible. By design, an audience measurement of websites and apps cannot see an AI called from business software, an internal tool or an automated agent. That kind of usage is growing in companies.
- It is a recent, France-only measurement. This is a first edition, covering the French population. Shares can move fast: follow the next editions rather than freezing your strategy on a single figure.
Is your site ready to be cited by AI? Get your free GEO score on 7 criteria, with a citation test on ChatGPT.
Analyze my site for free →Should you only work on your ChatGPT visibility?
No. ChatGPT should be your priority, but not your only target. The other AI engines reach specific audiences, rely on different sources and carry more weight in some professional uses.
Three reasons not to ignore Claude, Gemini and Perplexity:
- Their sources are different. ChatGPT, Gemini, Claude and Perplexity do not retrieve information the same way. A site can be cited by Perplexity and ignored by ChatGPT. We cover these mechanisms in our article on citation differences between Claude, Gemini and Perplexity. Working on your presence on Bing does not produce the same effects as working on your Google rankings, which Gemini relies on.
- Their audiences are not the same. An AI engine with 5% of total time can weigh much more in your target: developers, lawyers, B2B buyers, researchers. The barometer gives a national average, not the split within your industry.
- Part of their usage is invisible in this barometer. Built-in AI (Google, Microsoft) and API usage are not counted. The real weight of these ecosystems is therefore higher than their time share suggests.
→Perplexity: how to appear in its answers
Why weight your AI visibility by real usage?
Because being cited on a little-used AI engine is not worth the same as being cited on ChatGPT. A simple average across four engines gives each the same importance. An average weighted by time spent better reflects what your customers actually see.
Take a fictional example. A brand is cited in 20% of ChatGPT answers and in 60% of Claude, Gemini and Perplexity answers. With a simple average, its visibility looks good. But once you account for each engine's weight in time spent in France, the result is pulled down by ChatGPT, which weighs far more than the other three combined. The brand thinks it is visible while missing most conversations.
The reverse is also true. A brand that is highly visible on ChatGPT but absent elsewhere faces a less urgent but real risk: it depends on a single player.
That is the logic behind Beeleven's AI visibility audit, the agency that built Detekia. The audit asks ChatGPT, Gemini, Claude and Perplexity the questions your prospects really ask in your industry. It measures whether your brand is cited, in which position and against which competitors. Visibility is then weighted by the time spent on each AI engine in France, to reflect real usage rather than a theoretical average.
→GEO KPIs: what to measure and how
What should you do in practice?
Start with ChatGPT, check the fundamentals shared by every AI engine, then measure your visibility on the four major engines, weighted by usage. Six steps take you from intuition to steering.
- Make sure your site is readable by AI. AI bots allowed in robots.txt, content accessible without JavaScript, direct answers at the top of the page. The GEO score measures these fundamentals, shared by every AI engine.
- Prioritize ChatGPT. Understand how ChatGPT chooses its sources and work on your presence on Bing, which ChatGPT has long used for its web search.
- List your customers' questions. 20 to 50 real questions: comparisons, "best X for Y", problems you solve. These are what you should test, not just your brand name.
- Test these questions on the four engines. Note whether you are cited, in which position, and who is cited instead of you. The competitors ChatGPT cites are not always the ones you expect.
- Weight the results by usage. Use the Arcom time shares as a starting point, then adjust for your target (an engine that is little used on average may dominate in your industry).
- Measure over time. AI answers change, and so do their audiences. Repeat the measurement regularly and with each new edition of the barometer. Our guide to measuring AI visibility compares the available methods.
Step one: check for free whether your site is ready for AI. GEO score on 7 criteria and a citation test on ChatGPT, in under a minute.
Analyze my site for free →Frequently asked questions
What is the most used AI in France?
ChatGPT. According to the Arcom barometer, it was used by 27 million French people in April 2026 and accounts for 76% of the time spent on AI services. Claude (5%), Gemini (4%) and Perplexity (3%) are far behind.
Does the Arcom barometer include Google AI Overviews?
No. According to Arcom's presentation, the barometer covers dedicated AI websites and apps. Features natively built into devices or other services, such as AI answers inside Google Search, are outside its scope.
Should you give up on Claude, Gemini and Perplexity?
No. They carry little weight in average time spent, but they can carry a lot in your target audience. They also use different sources from ChatGPT. GEO fundamentals (structured content, cited sources, identified author) pay off on every AI engine anyway.
How do I know which AI engines cite my brand?
By asking each engine your customers' questions and analyzing the answers. The free Detekia score checks whether your site is ready to be cited and tests a few questions on an OpenAI model. For a complete measurement on ChatGPT, Gemini, Claude and Perplexity, weighted by real usage, see Beeleven's AI visibility audit.
Sources
- Arcom, "Baromètre d'audience des services d'intelligence artificielle, 1re édition", June 2026 (in French, Médiamétrie//NetRatings data).
- Arcom, announcement and presentation video of the barometer, June 16, 2026 (in French).
→Claude, Gemini, Perplexity: citation differences between AI engines