AI for Life
Asking an AI about your health: what it can do, and where it stops
People ask AI assistants about their health constantly, usually at eleven at night with a symptom and no appointment for three weeks. Pretending otherwise is useless. What is useful is knowing exactly where the line is: the publishers of these tools draw one themselves, and it is stricter than most users realise. Here is what is on each side.
The publishers draw the line before the law does
Start with the terms you accepted. OpenAI's usage policies list, among prohibited uses, tailored advice that requires a licence, such as legal or medical advice, without the involvement of a qualified professional. Anthropic's usage policy prohibits using its products to provide false or misleading information related to medical, health or science issues.
Read that first sentence carefully, because it is more precise than "do not trust the robot". What is forbidden is the tailored part: your symptoms, your history, your case, with a conclusion attached. General information about how a condition works is a different activity.
That distinction is the whole guide. An assistant can help you understand a subject. It cannot be your doctor, and its own publisher says so.
Why "but doctors use it" is not an argument
Health professionals do use AI systems, and in France the HAS and the CNIL have published a working document on doing it properly in care settings. Two of its points explain why their situation is not yours.
First, on liability: as of that document, no specific liability regime has been established for health professionals using AI. Their responsibility is assessed under ordinary law, which means they answer not only for the use they make of the AI but for the degree of autonomy they grant it in their acts and decisions. The document states it plainly: the health professional remains entirely responsible, even when the act is performed with the assistance of an AI system.
Second, on supervision: two criteria dedicated to AI systems were added to the sixth certification cycle for French healthcare establishments, applicable since September 2025. The professional operates inside a system of training, charters, patient information, human supervision and compliance.
When you ask a consumer chatbot about a mole, none of that exists. There is no professional carrying the responsibility, no certification, no supervision. That is the real difference, and it has nothing to do with how good the model is.
Note that this document is published as a working document, and that it addresses professionals and establishments rather than patients. We cite it for what it says about responsibility, which is the part that answers the question.
Four uses that are genuinely useful
All four have the same shape: they help you deal with the medical system, they do not replace it.
Prepare the consultation
The commonest reason a consultation goes badly is that the patient forgets half of it. This is a real use, and it involves no diagnosis:
I have a 15-minute appointment with a doctor about a recurring problem. Here are my notes. Help me prepare: turn this into a short factual summary of what happened and when, then list the questions I should ask. Do not suggest a diagnosis, a cause, or a treatment.
Translate the vocabulary, after the consultation
A discharge note or a prescription in professional language is a legitimate thing to want explained. Keep it impersonal:
Explain what this medical term means in plain language, in general terms: what it refers to, how it is usually investigated, what the words mean. Do not apply it to my case and do not tell me what I should do.
Understand a decision you were given
Asking better questions of your doctor is almost always more valuable than a second opinion from a chatbot. "What are the standard questions a patient should ask before accepting this type of examination?" produces a list you can take with you.
The paperwork
Appeals, reimbursement letters, forms, insurance correspondence: this is ordinary administrative writing where an assistant is good and the stakes are not clinical.
Your health data, in a tool that is not a medical record
Health information is among the most sensitive categories of personal data. A consumer assistant is not a medical record and gives you none of the protections one has. Our guide on protecting your personal data with an AI assistant covers the settings in detail. If you use it anyway:
- Stay impersonal. No name, no date of birth, no record number, no address. "A 40-year-old person" is enough for a general question.
- Do not upload the document as it is. Retype the term you want explained rather than photographing the whole report with the letterhead and your file number on it.
- Turn off history and data reuse in the privacy settings where the service allows it, before the first question.
- Never for someone else. Your parent's results or your child's are their data, not yours to paste.
Where to go instead, when it is urgent
An assistant has no ability to act. It cannot see you, call anyone, or escalate. In France, the official emergency numbers are listed by the public service: 15 for the Samu in a life-threatening medical emergency, 112 as the European emergency number, and 114 by message for deaf and hard of hearing people. Some regions also run 116 117 to reach a doctor on call outside surgery hours; check whether it covers yours. Outside France, look up your own country's numbers now, not during the emergency.
Three situations where you close the chat and contact a human: anything sudden or severe, anything involving a child or a pregnancy, and anything where you are frightened. Fear is not a clinical criterion, but it is an excellent reason to talk to a person.
The line, in one sentence
General information, preparation and paperwork: useful, and allowed. Your symptoms, your results, your decision: that is tailored medical advice, which the publishers of these tools prohibit without a qualified professional, and which nobody in the chat window is responsible for. We have run no hands-on test on any health-specific AI tool, so this guide names none and recommends none.
Tools in this guide
- Vendor information
ChatGPT
General-purpose assistant for drafting and research.
WritePrice: to verify
- Vendor information
Claude
Assistant for long documents and careful writing.
WritePrice: to verify
- Vendor information
Gemini
Google's assistant, linked to Workspace.
WritePrice: to verify
Questions
Can an AI assistant diagnose me?
No, and its publisher forbids it. OpenAI's usage policies prohibit tailored advice requiring a licence, such as medical advice, without the involvement of a qualified professional. Anthropic's usage policy prohibits providing false or misleading information on medical, health or science issues.
Doctors use AI, so why can't I?
Because the responsibility sits with them. In the HAS and CNIL working document on AI in care settings, no specific liability regime exists for health professionals using AI, and the professional remains fully responsible even when an act is performed with the assistance of an AI system. When you use a consumer assistant, nobody carries that responsibility.
Is it safe to type my symptoms into a chatbot?
Health information is among the most sensitive data there is, and a consumer assistant is not a medical record. If you use one, keep it impersonal: no name, no date of birth, no medical record number, and turn off history and data reuse where the service allows it.
Sources
- OpenAI, Politiques d'utilisation (usage policies)
- Anthropic, Usage Policy
- HAS et CNIL, Accompagner le bon usage des systèmes d'intelligence artificielle en contexte de soins, document de travail
- Service-Public.fr, Quels sont les numéros d'urgence (Samu, pompiers...) et les numéros d'écoute ?
One AI email a week. Only what changes something for you.