A patient survey, not a poll of all French people
On May 19, 2026, Doctolib and the Jean-Jaurès Foundation published the second edition of the France Maps of Access to Care, analyzed by health geographer Joy Raynaud. The study is based on 234 million consultations carried out in 2025 and, for the first time, on a survey of 7 845 patients. It is in this section that the figure that made headlines appears: 45 % of respondents say they have used a conversational AI, such as ChatGPT, for a health question in the past three months. Among them, 9 % only once, 22 % several times, and 13 % regularly.
A clarification is needed, which the study itself states: the sample consists of Doctolib users, weighted by gender and age. These patients are more familiar with digital tools than the average, which likely leads to overestimating AI usage compared to the general population. The figure of 45 % is therefore not that of "the French," but a strong signal for an already connected population.
A spectacular generational divide
Age is the most determining factor. 75 % of 18-24 year-olds and 66 % of 25-34 year-olds have used AI for their health, compared to 19 % of 65-74 year-olds and 13 % of 75 year-olds and over. The gap of 62 points is, according to the report, the largest in the entire survey. Joy Raynaud speaks, on ICI's microphone, of a "particularly strong generational divide." The professional category extends this finding: students (78 %) and executives (57 %) are the most frequent users, compared to 38 % of workers and 17 % of retirees, a gradient that, according to the study, raises the question of a digital divide in access to health information.
A territorial gap not to be overinterpreted
Usage reaches 49 % in dense urban areas, compared to 36 % in non-suburban rural areas. According to the authors, this gap mainly reflects composition effects, as cities concentrate more young people and executives, but it may also indicate lower digital access in rural areas. The study also suggests moving beyond the sole focus on medical deserts: the difference in healthcare avoidance between dense urban and rural areas is only 4 percentage points. It is the organization of supply, its availability and clarity, rather than medical density, that makes the difference.
When AI changes the care pathway
The survey asked the 3 527 respondents who used AI about its effects (multiple answers possible):
- 44 % state that AI did not influence their decision to consult;
- 25 % say they consulted more quickly, and 18 % consulted when they had not thought of doing so.
- 18 % say they gave up on consulting a professional after the exchange with the AI;
- 8 % first gave up before eventually having to consult, and 5 % went to the emergency room.
The report concludes that in more than 60 % of cases, the AI influences consultation behavior, in one direction or the other: accelerating for some, substituting for others.
Students, a high-risk profile
The most concerning finding concerns students. They have the highest AI adoption rate (78 %), the highest rate of forgoing care in the survey (66 % say they have already given up on seeking an appointment), the strongest financial barrier (37 % cite cost), and one of the highest rates of forgoing care after using AI (20 % of student users). The report summarizes:
“Faced with the cost and scarcity of appointments, the student turns to AI, which may reinforce a decision not to consult. This phenomenon calls for particular attention in terms of student health policy.”
The risk is clear: a generalist assistant does not perform a clinical examination, does not know the medical record, and can make mistakes with confidence. When it reinforces a decision to forgo care already motivated by cost or delays, it can delay a diagnosis.
What regulation?
The study argues for regulation of these tools and for solutions that guide toward the right care pathway. The European framework is beginning to change. Since 2 August 2026, Article 50 of the AI Act requires chatbots to indicate that they are not human. AI software falling under medical devices will be subject to “high-risk” rules, but the digital omnibus postponed their application to 2 August 2028; a generalist assistant asked about a symptom does not, in principle, fall into this category. Finally, since its designation under the DSA on 31 August 2026, ChatGPT must assess by the end of December 2026 the systemic risks of its service, including for the physical and mental well-being of its users.
Regulation of the tools will not be enough. The survey reminds us that 63 % of patients have already given up on seeking an appointment and that 23 % went to the emergency room due to lack of availability. If some young people turn to AI, it is also because access to a doctor is seen as costly and difficult. Finally, note that Doctolib itself is a digital health player, which launched a research project in AI in summer 2026 based on its users' data: a reason to read these figures with the critical attention that any study produced by a stakeholder warrants.
Sources: Doctolib and Jean-Jaurès Foundation, “Maps of France 2026 of access to care” (full study); Doctolib, presentation of the study, 19 May 2026; Jean-Jaurès Foundation; HERE, 19 May 2026; European Commission, application of the AI Act (IP/26/1714); EU Council, omnibus on AI, 29 June 2026; European Commission, designation of ChatGPT under the DSA; franceinfo, Doctolib research project, 16 Jul. 2026.
