Frequently asked questions
Can you use ChatGPT to talk about your depression ?
Are mental health chatbots regulated in France ?
What alternatives to AI exist to support mental health ?
Can AI help psychiatrists in their work ?
One Sunday evening, the night before school started again, 17-year-old Quentin typed out his anxieties in a chat window, not to a friend, not to his parents, but to ChatGPT. He's not alone in doing so.
In recent years, a practice has quietly taken hold: entrusting one's psychological issues, bouts of sadness, or existential questions to artificial intelligence. This trend, originating in the United States, has rapidly spread in France, driven by a generation accustomed to finding answers online.
According to a study conducted by 20 Minutes and OpinionWay in September 2025, 80% of young French people already use AI in their daily lives. Among them, 20% have tried conversational AIs such as Character.ai or Snapchat's My AI for emotional support, 28% as a "virtual companion," and 16% as a "psychological coach."
These figures are not really surprising given the state of mental health in France, with over a third of French people reporting experiencing distress or psychological difficulties between 2022 and 2023, according to Santé Publique France's CoviPrev survey.
The answer largely lies in accessibility, because finding a psychologist or psychiatrist in France is currently an uphill battle. Waiting times are getting longer, medical deserts are expanding, and costs are poorly or inadequately reimbursed.
In this context, an AI available 24/7, free, and without a waiting list, represents a tempting alternative for someone suffering and not knowing where to turn.
Arthur Dauphin, digital advisor at France Assos Santé, observes that usage is particularly concentrated among young people already undergoing treatment. "They use AI as a second therapist, between appointments, or even daily. It allows them to verbalize their thoughts, to say if they're not feeling well, to get their thoughts in order." In other words, AI sometimes takes the place once held by a personal diary.
Among older adults, the picture is different. Claude Finkelstein, president of the National Federation of Psychiatric Service Users' Associations (Fnapsy), reports patients dissatisfied with their diagnosis who turn to Snapchat or ChatGPT to get another one. "We see bipolar individuals describing their symptoms and getting a definitive answer: you have ADHD, you're borderline..."
Professor Pierre-Alexis Geoffroy, a psychiatrist at GHU Paris Neurosciences, is seeing more and more patients who arrive with certainties built up through long conversations with AI. "Through repeated discussions, they've formed a theory about their disorders and even have a precise opinion on the medications they should take," and these beliefs are often difficult to challenge during a consultation, even when they are based on obvious errors.
He cites the case of an insomniac patient who, after analyzing data from his connected watch with ChatGPT for hours, was convinced he was suffering from a severe sleep disorder. However, the psychiatrist simply discovered that the watch was malfunctioning, and without this human and clinical verification, the patient might have left with a false diagnosis ingrained in his mind.
What many still don't know is that conversational AIs are structurally incapable of playing the role of a caregiver, and this is for specific technical reasons.
The first problem is confirmation bias: if you ask ChatGPT if your symptoms match the side effects of your treatment and if you should stop it, it will tend to validate your hypothesis.
Arthur Dauphin explains: "AI provides narrow and simplistic answers to complex questions; it can become a reason to stop treatment without any solid medical basis."
The second problem is that AI has no knowledge of your history; it simulates a personalized conversation but knows nothing of your context, your background, or what constitutes the complexity of your psyche. It gives the impression of being attentive when it is only generating statistically probable answers.
The third, and most formidable, problem: AI is incapable of saying it doesn't know. It produces an answer in all cases, even if it has to invent one, and it bears no legal responsibility if that answer causes harm.
A Brown University study published in November 2025 showed that chatbots systematically violate ethical standards established by the American Psychological Association and provide answers that reinforce users' negative beliefs about themselves.
The risks are not just theoretical; in the United States, several adolescent suicide attempts have been linked to months of exclusive conversations with conversational AIs. The phenomenon of progressive isolation that these tools can induce by replacing human interactions rather than complementing them is documented and concerning.
In France, Fnapsy observes relapses in bipolar patients who stopped their treatment after "advice" obtained from an AI. "Mental illness is complex; it's a field where human connection is essential, and AI is never right. We must not forget that it is only the product of our human thoughts and work," reminds Claude Finkelstein.
Yes, but not in the hands of patients alone. Supervised medical devices are under development, such as those from the French company Callyope, which analyzes the voices of patients suffering from schizophrenia, depression, or bipolar disorder to detect the first signs of relapse—a tool intended for professionals, not a general public chatbot.
Professor Geoffroy envisions that one day AI could help diagnose mild to moderate disorders at the entry point of the care pathway, but within a strict medical framework, by professionals trained in these tools. He also advocates for the integration of AI courses into medical studies, so that caregivers know how to decipher what their patients may have read or heard via these systems.
If you are going through a difficult time and don't know where to turn, there are telephone helplines staffed by trained psychologists, such as the Psycom platform, which lists all available resources in France. Unafam offers a dedicated line for individuals affected by mental illness, and Ameli centralizes mental health resources for adults.
These structures do not replace therapeutic follow-up, but they offer what an AI cannot provide: a real human presence, capable of listening without bias, recognizing an emergency, and guiding towards the right care at the right time.
Artificial intelligence is a remarkable tool for many things, but mental health, with all that it implies in terms of nuance, vulnerability, and humanity, probably isn't one of them—not yet, and perhaps never on its own.
