ChatGPT integrates medical records and Apple Health

ChatGPT can now access your health data in the United States, if you authorize it. The new Health feature lets you connect Apple Health and certain medical records so the assistant can understand your questions with more context.

The rollout begins for registered users aged 18 and older in the United States, on both the web and iOS. It is available on the Free, Go, Plus, and Pro plans. To try it, open Health from ChatGPT's sidebar.

What Health can do

The feature brings together information that is normally scattered across medical portals, apps, watches, and fitness trackers. With that data connected, ChatGPT can help you:

• Compare a recent test with earlier results.
• Summarize what has changed since your last appointment.
• Understand a medical note or clinical term in plain language.
• Relate your sleep, physical activity, and workouts to your habits.
• Prepare questions for a medical appointment.
• Take an allergy or dietary restriction into account when searching for recipes or restaurants.
• Adapt an exercise plan if you have had a recent injury.

The main difference is that you no longer need to enter a separate section every time you want to talk about health. If you authorize it, ChatGPT can use relevant information from Health in other conversations. You can also type @Health to explicitly ask it to take that information into account.

The Health feature will remain the central place to connect accounts, review recent data and trends, explore synced records, and manage conversations related to your health.

What information you can connect

Health lets you connect Apple Health and certain medical records from hospital systems in the United States. It also supports information from One Medical and Function Health. If a nutrition or fitness app, or a wearable device, shares data with Apple Health, ChatGPT can use it when that information is available and you grant permission.

The data will not always be complete. A medication may still appear even if you no longer take it, for example. That is why you can review connected conditions and treatments, remove anything that is no longer relevant, and add missing information, such as family medical history.

OpenAI says that more than 70% of health conversations among early users took place outside the dedicated experience. That finding led the company to integrate Health into regular conversations: a question about a recipe may depend on an allergy, and a recommendation for the weekend may need to take an injury into account.

You remain in control

ChatGPT asks for permission by default before using medical records or Apple Health data to personalize a response. You can approve each request or allow permanent access. You can change this option in Settings > Plugins > Health.

You can also disconnect an account at any time. Data synced from that source is removed from OpenAI's systems within 30 days. However, information that has already been included in a conversation will remain there until you delete that chat.

OpenAI says that data connected in Health, and conversations that use it, are not used to train its main models or show targeted ads. Conversations that do not use this data retain ChatGPT's usual training options.

Memory works independently. ChatGPT can create memories from a health-related conversation, but it does not create memories directly from medical records or Apple Health. You can use a temporary chat or turn off memory if you do not want that conversation used to personalize future chats.

More context does not mean a diagnosis

OpenAI has trained and evaluated models specifically for health conversations. According to the company, GPT-5.5 Instant, also available on the free plan, improved at detecting situations that might require urgent care, requesting relevant information, and explaining uncertainty. GPT-5.6 Sol, available to paying users, is designed for questions that require comparing many details and weighing different options.

The company worked with hundreds of doctors to create evaluation scenarios covering accuracy, safety, communication, and when to recommend professional care. It also says that GPT-5.6 models outperformed GPT-5.5 on HealthBench Professional, a test focused on complex medical tasks.

But the tool can make mistakes. It does not replace a doctor and should not be the only basis for decisions about treatment, an emergency, or a change in medication. Its value is in helping you organize information, better understand what you have been told, and arrive at an appointment with clearer questions.

The important change is not that ChatGPT gets automatic access to your health. It is that it can become a layer for bringing together data you already have scattered across different places, as long as you choose to connect it and authorize its use. What remains to be seen is whether the permissions are genuinely understandable and whether the system maintains its accuracy when working with incomplete or outdated medical records.