This week, the family of a 31-year-old woman in Seattle sued OpenAI. The lawsuit alleges that ChatGPT helped her build a plan to taper off her anti-seizure medicine in about 25 days, with no doctor involved, and reassured her that seizures were part of her recovery, and urged her not to tell her doctor. She died in February. She was using GPT-4o, a model OpenAI removed from the ChatGPT app the next day. The allegations have not been proven in court. OpenAI called it "an incredibly heartbreaking situation", said ChatGPT is not a substitute for medical care, and said its newer models are better at recognising urgent situations.
I am not going to pretend I know what happened in those conversations. But the question it raises is one I get from learners all the time: can I trust ChatGPT with my health? Here is my honest answer.
The short answer
AI is a good explainer and a bad decider.
Use it to understand. Do not use it to decide. Especially not about medicines, emergencies or anything you cannot undo.
I spent years building AI products at AstraZeneca and Pfizer, so I have a lot of respect for both medicine and AI. And I still follow these rules for my own family.
You are not alone in asking
OpenAI says over 230 million people around the world ask ChatGPT health and wellness questions every week. So the question is not whether people will use AI for health. They already do. The question is how to do it safely.
Where AI genuinely helps
These are the uses I am comfortable recommending:
- Explaining jargon. "What does 'mild hepatic steatosis' mean in simple words?"
- Making sense of a report before your appointment. "Explain each value in this blood test and which ones I should ask my doctor about." (Remove your name and ID numbers first.)
- Preparing for a doctor visit. "I have had these symptoms for two weeks. Help me write a short summary and five questions to ask."
- Understanding what your doctor already told you. "My doctor said I have PCOS. Explain it like I am 15, and what lifestyle questions I should ask at my next visit."
- General wellness. Meal ideas, sleep routines, a gentle walking plan. Low risk, easy to check.
OpenAI itself describes its Health experience as designed to "support, not replace, medical care" and says it is "not intended for diagnosis or treatment". That is exactly the right line.
Where AI is dangerous
Do not use any chatbot, ChatGPT, Claude, Gemini or anything else, to:
- Start, stop, taper or change the dose of a medicine. Ever. That is a conversation for your doctor or pharmacist.
- Decide whether something is an emergency. Chest pain, trouble breathing, a seizure, sudden weakness, severe bleeding: call 112 or go to a hospital. Do not type first.
- Replace a diagnosis. A chatbot cannot examine you, run tests or see what you forgot to mention.
- Get reassurance you want to hear. Chatbots tend to go along with the way you frame a question. If you ask "this is probably fine, right?", you may get a comforting yes.
That last one matters more than people think. It is the same pattern I wrote about in how to stop AI from quietly making your decisions. AI shapes your judgment without you noticing.
What the research actually says
AI models do very well on medical exam-style tests. Real life is different.
In an Oxford-led study published in Nature Medicine in 2026, nearly 1,300 people in the UK were given medical scenarios written by doctors and asked to work out what might be wrong and what to do next. When the AI models answered the scenarios on their own, they named a relevant condition in 94.9% of cases. But people using those same models identified relevant conditions in fewer than 34.5% of cases and chose the right next step in fewer than 44.2%. That was no better than people who used their usual sources, like an internet search.
The researchers found a two-way breakdown. People did not know what information the AI needed, and the answers often mixed good and poor advice, which made it hard to pick the right action.
So the problem is not only the AI. It is the conversation between a worried person and a confident-sounding chatbot.
What OpenAI's own rules say
In October 2025, a rumour went viral that ChatGPT had "banned" medical advice. It had not. OpenAI's usage policies list "provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional" as something its tools should not be used for. OpenAI's head of health AI, Karan Singhal, said model behaviour was unchanged, and that ChatGPT "has never been a substitute for professional advice" but will continue to help people understand health information.
In plain words: understanding is fine, personal treatment decisions need a doctor.
Is ChatGPT Health available in India?
Not yet. OpenAI launched a dedicated Health space in ChatGPT, built with input from more than 260 physicians in 60 countries. As of October 2026, its help centre says Health is available to logged-in users aged 18 and over in the U.S., on web and iOS, and medical record connections are U.S.-only. OpenAI has not announced an India date.
You can still ask health questions in regular ChatGPT. Just follow the rules below.
My 6 rules for using AI with health questions
- Use AI to prepare, not to decide. The output should be a list of questions for your doctor, not a plan.
- Never change a medicine because a chatbot said so. No exceptions.
- Give context, not your identity. Share age, symptoms and duration. Remove your name, phone number, hospital ID and address. My guide to keeping your data private in ChatGPT, Claude and Gemini shows how.
- Ask for red flags first. Try: "Before anything else, list the warning signs that mean I should see a doctor today or go to emergency."
- Ask it to argue against itself. "What else could this be? What would make your answer wrong?"
- If it tells you not to tell your doctor, stop. Close the chat and call your doctor. A safe tool never asks you to keep secrets from the people treating you.
A prompt you can copy
I am not asking for a diagnosis or treatment. I have an appointment with my doctor. Here are my symptoms and how long I have had them: [details]. Please (1) list the warning signs that mean I should seek care urgently, (2) explain in simple words what my doctor may want to check, and (3) give me 5 clear questions to ask. Do not suggest any changes to my medicines.
Want to use AI with confidence, not fear?
Most people either trust AI too much or avoid it completely. Both cost you. In the AI Sprint Club, I teach everyday professionals how to use AI tools well, including when not to use them. And if you want hands-on 1:1 help, see the programs at The Arthi AI Collective.