OpenAI opens a new phase in the relationship between artificial intelligence and personal health. It is called Health ChatGPT and is a dedicated, platform-integrated space designed to help people navigate clinical documents, wellness apps, and health information. It does not promise diagnoses or therapies, and indeed makes it clear up front what it will not do. The stated goal is more pragmatic: to make often fragmented data understandable and better prepare for dialogue with health professionals.
An assistant
OpenAI insists on one point: ChatGPT Health “is designed to support medical care, not replace it.” The system is intended to support people with everyday questions, interpretation of exams, and preparation for visits. It is a role that intercepts a habit that is already widespread: according to the company, more than 230 million people worldwide ask ChatGPT health and wellness questions each week, based on anonymous analysis of conversations. Hence the idea of a dedicated space with stricter rules.
Scattered data, a single vision
The main promise concerns the organization of information. Today, reports, apps, wearable devices, and pdfs live on different platforms. ChatGPT Health makes it possible to upload documents and link apps such as Apple Health, MyFitnessPal or Function, so conversations are based on user-selected data. In the United States, it is also possible to link electronic health records through dedicated partners. AI can help understand trends in parameters over time, summarize recent exams, or suggest targeted questions for a visit.
Privacy: a separate space
The most sensitive issue is data protection. OpenAI has created an isolated space within ChatGPT: health-related conversations, files, and apps remain separate from other chats. Health information is not used to train basic models. Added to this are dedicated encryption, isolation mechanisms, and the ability to delete memories and files at any time. The company emphasizes that control always remains with the user, including immediate revocation of access to connected apps.
Two years of comparison
The development took two years and involved more than 260 physicians in 60 countries and numerous specialties. Their input was not advisory: they evaluated the model’s responses more than 600,000 times, influencing how AI communicates, when it suggests referral to a physician, and how it handles potentially risky cases. This approach flowed into an evaluation system called HealthBench, which prioritizes safety, clarity, and proper clinical context.
Gradual access
ChatGPT Health is available initially to a small group of users, with access via a waiting list. Expansion to all on web and iOS is expected in the coming weeks. More advanced integrations, such as electronic health records, remain limited to the United States for now.
The entry of AI into personal health data management is a structured experiment, with stated limitations and a focus on security. Its real usefulness will depend on how it is used: as a tool for guidance and awareness, or as an improper shortcut in a field that does not admit of simplification?
