Connect your medical records to an AI? A decision and a setup check

Short answer

Connecting your records can be worth it for one clear job: explaining test results in plain words, putting your history in order before a visit, or drafting questions for your clinician. If what you want is a diagnosis or a decision about treatment, stop: the app itself says it is not for that. In the US, the privacy law that binds your clinic usually does not cover an app you choose yourself, so the app's own terms are your protection. If you go ahead, connect the least you need, lock the account, check every figure against the original, and set a date to review or disconnect.

Who this is for. Adults, carers and people with a long-term condition who have been offered a "connect your records" button. Needs 15 minutes and the app on your phone or computer.

What is on offer (as of October 2026)

OpenAI launched ChatGPT Health on 7 January 2026, and a 23 July 2026 update says it is "launching to U.S. users 18 and older on web and iOS" (OpenAI, vendor). It connects medical records through b.well (a company that links US healthcare providers) and apps such as Apple Health, Function, MyFitnessPal, Weight Watchers, AllTrails, Instacart and Peloton. "Medical record integrations and some apps are available in the U.S. only," and Apple Health needs an iPhone (same source).

What OpenAI says about your data, as of October 2026:

• Training. "Connected medical records and Apple Health information, and conversations that use it, are not used to train our foundation models or target ads" (OpenAI Help, vendor help).
• Purpose. Health "is not intended for diagnosis or treatment" (same source).
• Where your data shows up. The January announcement said "Health has separate memories" (OpenAI, vendor). The current help page says something broader: "Once your data is synced in Health, you can ask questions using that data anywhere in ChatGPT," and "When memory is on, conversations that use Health can create memories to personalize future chats." Only the older Health Project keeps its memories "contained to that project" (OpenAI Help, vendor help). So do not assume your health data stays in one box.
• Disconnecting. "When you disconnect an account, data synced from that source is deleted from OpenAI's systems within 30 days" (same source). Chats you already had stay until you delete them.

Other assistants may add similar features. The questions below work for any of them. If your question is about AI used by your clinic, such as a listening scribe or AI-drafted portal replies, see AI at the doctor's.

Step 1: the yes or no check

Know the protection gap. In the US, HIPAA (the federal health privacy law) covers your doctor's office, your health plan and the companies working for them, but an app you choose and use yourself is usually not covered (HHS, 2016, regulator guidance). Many such apps fall under the FTC's Health Breach Notification Rule instead, which makes them tell you after a breach of your health information (FTC, 2024, regulator guidance). A rule about telling you after a leak is not the same as a rule about what they may do with your data.

Step 2: a six-item setup check

Our suggestion, in this order.

1. Connect one source. Your main clinic's records, or one app. Not every app on the list. Add another only when a job needs it.
2. Share the fewest data types. For Apple Health, OpenAI's help page says you choose the categories to share (Health, then "...", then Apple Health, as of October 2026). Leave out any category the job does not need.
3. Lock the account. Turn on multi-factor sign-in (a code or passkey as well as your password); OpenAI's announcement suggests it (OpenAI, vendor). Sign out on shared computers, and check that no one else uses your login.
4. Decide about memory. If you don't want health chats to shape other chats, use Temporary Chat for health questions, or turn memory off in Settings, then Personalization (as of October 2026). AI memory: what it keeps has a ten-minute audit.
5. Take out what the job doesn't need. When you type or upload, the usual rule applies: What not to paste.
6. Put a review date in your calendar. Three months is our suggestion. On that day, run the disconnect routine below or decide to keep it.

Step 3: use it, then check it

Ask for explanations and questions, not verdicts. Then check every number, date, medicine and dose it mentions against the original record or with your clinician. A connected record makes the answer sound more personal. It does not make it more correct.

Using only my connected <SOURCE> records from <DATE_RANGE>:
1. List each test result outside its reference range, quoting the value and range exactly.
2. Explain each term in plain words.
3. Do not diagnose me or suggest a treatment.
4. Write five questions to ask my clinician, most important first.

Replace the following:

• <SOURCE>: such as clinic or Apple Health.
• <DATE_RANGE>: such as the last six months.

Disconnect and delete

As of October 2026, from OpenAI's help page (OpenAI Help, vendor help):

1. Medical records: Health, then Accounts, then "..." next to the account, then Remove account.
2. Apple Health: Health, then "...", then Apple Health, then Disconnect.
3. Old chats: delete the conversations that used your health data. Disconnecting does not remove them.
4. Memories: OpenAI's announcement says Health memories can be viewed or deleted in Health or in Settings, then Personalization (OpenAI, vendor). Delete any you don't want kept.
5. Write down the date. OpenAI says synced data is deleted within 30 days of disconnecting.

Limits

Sources

1. OpenAI. Introducing ChatGPT Health, 7 January 2026, updated 23 July 2026. Vendor announcement. Link
2. OpenAI. Health in ChatGPT, help article, read October 2026. Vendor help. Link
3. HHS Office for Civil Rights. Health App Use Scenarios and HIPAA, February 2016. Regulator guidance. Link
4. FTC. Health Breach Notification Rule: The Basics for Business, July 2024. Regulator guidance. Link

Changes

1. v1.019 Oct 2026first version