ChatGPT Health is open across the US. Treat it as an organizer, not a doctor

Who can use it and where

OpenAI's help material says Health is available to eligible logged-in Free, Go, Plus and Pro users in the United States who are 18 or older. It is supported on the web and on iOS versions of ChatGPT 1.2026.188 or later. That scope should not be generalized to children, every country or every device.³

The company's July update says the US launch follows an earlier limited test. TechCrunch reported that users can connect records from participating health systems and data from services such as One Medical and Function Health. Availability of a specific provider still depends on the integration, account and region.^1,2

A dedicated Health space separates this use from an ordinary conversation in the interface. That can make controls easier to understand, but it does not make every connected record complete. A hospital portal may omit outside care, old scans, handwritten notes or a medication change made elsewhere.^1,3

• US users must be 18 or older and signed in.
• Access covers listed plans on web and supported iOS versions.
• A connected source may still be an incomplete medical record.

What it is useful for

The strongest tasks are administrative and explanatory. Ask for a timeline of lab results, a list of medication changes, or plain-language definitions of terms in a discharge note. Then compare the answer with the original documents before relying on it.^1,3

It can also help prepare for an appointment. A person with several records can ask for a short list of unresolved questions, dates and trends to discuss. That may reduce time spent searching during a visit without asking the model to decide what the symptoms mean.¹

Tracking patterns can be useful when the inputs are consistent. Sleep, exercise, nutrition and symptom notes may reveal dates worth discussing. Correlation is not diagnosis. A pattern that appears in an app may reflect missing days, a changed sensor or a measurement error rather than a health change.^1,2

• Use it to summarize, organize and prepare questions.
• Keep the original record open while checking the summary.
• Label self-tracked data separately from clinician-entered data.

Privacy deserves a deliberate setup

Health information is among the most sensitive data a person can share. Before connecting anything, read which accounts will be linked, what categories will be imported, how to disconnect them and how to delete data from the Health area. Save a copy of important records outside the chatbot.^1,3

OpenAI describes privacy controls specific to Health and says its design was informed by clinicians. Those are company claims readers can inspect in the source material, not a reason to skip personal risk assessment. The right setup for a routine fitness log may be wrong for psychiatric notes, reproductive health or a rare diagnosis.^1,3

Consider using the minimum data needed for the task. If the question concerns one lab panel, uploading years of unrelated records may add risk without improving the answer. Redact names, addresses or family details when they are irrelevant and the source format allows it.^1,2

• Connect only the record sources you intend to use.
• Review deletion and disconnection controls before importing.
• Share the minimum information needed for a specific task.

Where the medical boundary sits

Do not use a chatbot to decide whether urgent symptoms can wait. Chest pain, trouble breathing, signs of stroke, severe allergic reaction, suicidal thoughts and other emergencies need local emergency services or qualified urgent care, not another prompt.²

For non-emergency questions, carry the generated summary to a clinician and ask what is missing or wrong. If the answer recommends starting, stopping or changing a medicine, treat that as a question for the prescriber. Drug interactions and individual history are exactly where incomplete context can cause harm.^1,2

The expansion makes ChatGPT Health easier to try, but access is not the same as validation for every medical use. The product can lower the clerical burden around care. It should not lower the threshold for getting human care when the consequences of an error are high.^1,2,3

Families should decide whose account and records are being used. A caregiver may help organize information for another adult, but authorization, account access and the person's wishes still matter. Avoid mixing several people's records in one thread, because a later summary can attribute a medicine or diagnosis to the wrong person.^1,3

Export a clinician-ready summary as a draft, then edit it for accuracy and relevance. Remove speculative model language, distinguish symptoms from diagnoses and put dates beside measurements. A short corrected timeline is safer and more useful than handing a clinician a long chatbot transcript that mixes records, questions and generated interpretation.^1,2,3

• Never delay emergency care to continue a chatbot conversation.
• Confirm medication and treatment changes with a qualified clinician.
• Bring summaries back to the original records and care team.