OpenAI has relaunched Health in ChatGPT, rolling out to all logged-in US users aged 18 and older on Free, Go, Plus, and Pro plans. The feature had been in quiet beta since January because early testing found it giving wildly inconsistent results when you asked the same question, and a tendency to forget basic information like a user’s age and gender despite having full access to their records.
The relaunch is available on iOS and web. Once enabled, the feature can pull from connected Apple Health data and medical records to compare lab results over time, summarize changes since a prior appointment, track medications, and surface patterns across sleep, activity, and workout data. It also feeds into general ChatGPT conversations: the feature can factor in a dietary restriction when recommending a restaurant, or flag a recent injury when you ask it to suggest weekend activities. ChatGPT will ask for permission before pulling health data into any response.
OpenAI is keeping health conversations in a dedicated section within ChatGPT, separate from the main chat history. Connected medical records and Apple Health data will not be used to train OpenAI’s models, and they will not be used to target ads. What OpenAI does not claim is HIPAA compliance, which means this is not a covered health service in the legal sense and carries none of those protections.
Users who would rather not connect records at all can still use the Health section to ask general health questions. The connected data is strictly opt-in, and permission is requested per-conversation before health context is applied.
OpenAI built this feature over two years with input from more than 260 physicians across 60 countries, which makes the January accuracy failures particularly stark. The company announced the feature in January 2026 but never pushed it wide, almost certainly because the early results were not good enough to defend publicly. The relaunch presumably reflects improvements to OpenAI’s underlying models since the start of 2026, though the company has not detailed what changed in the health feature itself.
This tool is useful for spotting patterns and preparing questions for a doctor, not for drawing clinical conclusions. The inconsistency problems from January were serious enough that anyone using this to interpret their own bloodwork or health data should treat the output as a starting point, not an answer. Until independent testing validates the claimed improvements, treating early results with skepticism is the safer approach.
