ChatGPT Health can connect Apple Health and medical records. Connecting is optional — and still not a doctor

On July 23, 2026, OpenAI began rolling out Health in ChatGPT to logged-in U.S. users age 18 and older on web and iOS, across Free, Go, Plus, and Pro plans. You can optionally connect Apple Health and supported medical records so ChatGPT can use that context — with your permission — when you ask about labs, appointments, sleep, activity, or everyday health questions.

We read OpenAI's July 23 launch post in full, checked OpenAI's Help Center page for Health in ChatGPT, and cross-checked availability and connector details against Fierce Healthcare's July 24 coverage.

If you already ask ChatGPT to plain-English a lab result, prep questions for a doctor visit, or make sense of a portal note, this matters as a control decision: whether to keep pasting documents yourself, or to let OpenAI hold a connected stream of health data under the rules it published.

Company product announcements are the highest gap-risk source type in this category. OpenAI is describing its own product. The durable facts for readers are who can use it, what can connect, what OpenAI says it will not do with the data, and the limit that never changes: this is not medical care.

What actually shipped, per OpenAI: Health lives in the sidebar or More menu on web and iOS for eligible U.S. adults. Connectors include Apple Health, medical records from participating U.S. hospital systems, One Medical, and Function Health. Wearables and fitness apps that already share into Apple Health can flow through that pipe when you allow it. Availability varies by provider and metric; some proprietary scores may not transfer. Health is not yet in Codex or Voice mode. OpenAI's own scale claim: more than 300 million people turn to ChatGPT with health-related questions each week. Fierce Healthcare's July 24 write-up adds that supported records can include systems tied to Epic and Oracle Health — how this reaches some ordinary hospital portals, and stays unavailable for others.

The product shape shifted after an earlier limited experiment that kept Health more separate. OpenAI says more than 70% of health-related conversations among early testers happened outside that dedicated space. The July rollout leans in: once you connect data, you can allow ChatGPT to use relevant connected information across ordinary chats. By default it asks first; you can approve once, always allow, or change Settings > Plugins > Health later. Type @Health when you want the context on purpose.

On privacy, the published rules are specific. Connected medical records and Apple Health information, and conversations that use them, are not used to train foundation models or target ads — OpenAI says this holds even if your ordinary ChatGPT training toggle is on. Extra encryption applies to Health connections. Disconnect in Health > Accounts; OpenAI says synced data from a disconnected source is deleted from its systems within 30 days, while text already in chat history stays until you delete those chats. Memories may still form from Health conversations, not directly from the connected feed. Temporary Chat or turning memory off is the quieter path when you want less residue.

Model-score claims — GPT-5.5 Instant, GPT-5.6 Sol, physician evals, HealthBench — are OpenAI's own performance story: useful as "they are investing here," weak as proof your next portal note will be interpreted perfectly. OpenAI itself says ChatGPT can still make mistakes and does not replace clinicians. Its Help Center also says Health is not intended for diagnosis or treatment.

The significance assessment, measured against this reader's life: more useful than a gimmick, less automatic than "ChatGPT knows your chart now." If you advocate in exam rooms, juggle specialists, or keep a folder of PDFs for a parent, a carefully permissioned connection can save re-explaining the same history. If you are not ready to put clinical data in a consumer chatbot — trust, shared device, legal, or gut — you lose nothing by declining. Pasting one result into Temporary Chat remains a valid lower-commitment path. The adult decision is optional, reversible, and separate from ordinary non-health ChatGPT use.

One honest reaction: part of me is relieved the training-and-ads carve-out is stated in hard language. Another part stays wary that "not used for training" is not the same as "never stored, never mis-summarized in a way that embarrasses you in a real appointment." Relief and wariness can sit in the same chair.

This is the situation as of August 2, 2026. Connectors, defaults, and availability can change after launch weeks. Check Health in your own account and OpenAI's current Help pages before acting on any specific claim.

If you use ChatGPT on an iPhone and already live in Apple Health, open Health from the ChatGPT sidebar once and read the connect screens before you tap anything — especially the permission default and the disconnect path. Before you link a full medical record for yourself or a family member you help manage, decide the job: "explain this one lab," "timeline my last year of visits," or "help me write questions for Tuesday." If the job is one document, you may not need a standing connection at all.

A chatbot that can see your chart is still a chatbot. The grown-up feature is the off switch.