ChatGPT Health Privacy Checklist: What to Review Before Connecting Records

Decide whether connection is necessary

ChatGPT Health can combine information a person chooses to connect, including supported medical records and Apple Health, so an answer can refer to personal context. That can reduce repetitive uploading and help organize a timeline. It also creates a larger, more sensitive data flow than asking a general health question without records. Begin with a purpose: comparing two lab dates, preparing questions for an appointment, or reviewing a sleep trend. If the task can be answered with a redacted value or a short summary, a full account connection may be unnecessary.

Data minimization is a practical safety control, not merely a privacy slogan. Connect the fewest sources, for the shortest useful period, and avoid importing unrelated history. Confirm identity and dates inside the record because duplicate patients, delayed synchronization, inconsistent units, and stale medication lists can produce a polished but misleading synthesis. The product documentation describes grounded responses; it does not guarantee that every record is complete or interpreted correctly.

• Purpose before connection
• Minimum useful source
• Verified dates and units
• Defined review and disconnection point

Understand the controls as separate layers

OpenAI says connected medical records and Apple Health information, and conversations that use them, are not used to train foundation models or target ads. The same documentation says ordinary conversations that do not use connected Health data follow the account's standard model-training choice. Users should therefore avoid treating one Health promise as a description of every chat, file, connector, or account setting.

Permission controls govern when connected data may personalize an answer. OpenAI says the default is to ask before use, with an option to always allow it. Leave prompts on while learning the workflow; the extra confirmation makes unexpected use visible. Memory is another layer. Current help documentation says conversations using Health can create memories when memory is enabled, though memories are not created directly from synced Health data. Temporary Chat or disabling memory can be appropriate for a one-off sensitive question.

Account security protects all of those settings. Use unique credentials, multi-factor authentication, protected recovery channels, and session review. Avoid shared accounts. Before approving an action through another connected app, inspect the recipient and exact information being sent; a correct answer can still become a privacy incident through an overly broad share.

Plan deletion before importing data

Disconnecting and deleting are related but not identical. OpenAI's current help article says that disconnecting a medical-record or Apple Health source causes synced data from that source to be deleted from OpenAI systems within 30 days. It also says information already included in ChatGPT conversation history remains until the user deletes those conversations. A departure checklist should therefore cover connected accounts, conversations, uploaded files, memories, shared links, exports, and any downstream service that received information.

Read the Health Privacy Notice rather than relying only on a feature summary. It describes collected data, operational uses, service-provider processing, disclosures, controls, and consumer-health rights. Policies and interfaces change, so save the review date and repeat it after a material product update. This article summarizes opened documentation as of August 1, 2026; it is not a security audit or hands-on test of ChatGPT Health.

Separate privacy promises from legal coverage

HIPAA does not cover every product that handles health-related information. HHS provides a mobile-health-app tool because the answer depends on the product's functions, the data it collects, and whether the developer acts for a HIPAA covered entity or business associate. Direct-to-consumer services may instead or additionally face the FTC Act, the FTC Health Breach Notification Rule, state consumer-health laws, and other requirements. Individuals should not assume that a familiar medical logo or record connection transfers a provider's complete legal framework to every downstream service.

The FTC explains that its Health Breach Notification Rule can apply to vendors of personal health records and related entities outside HIPAA. A covered breach is not limited to hacking: unauthorized disclosure can also trigger review and notification duties. Organizations integrating AI with health data should map every source, processor, recipient, authorization, contract, and breach-response owner with counsel. Consumers can use the same map in simpler form: who supplied the data, who receives it, for what purpose, how long it remains, and how to revoke access.

Use Health to prepare for care, not replace it

OpenAI says ChatGPT Health is designed to support rather than replace medical care and is not intended for diagnosis or treatment. A safe workflow uses it to translate terminology, assemble a timeline, identify questions, and compare a generated summary with the underlying record. It does not let the model change medication, dismiss a symptom, interpret an emergency, or make a consequential decision without a qualified professional.

Ask the system to cite the date and source for every material statement, distinguish recorded fact from inference, list missing context, and show conflicting values. Then open the original record. For lab results, confirm units and reference ranges; for medication, confirm dose, route, frequency, prescriber, and current status; for trends, check whether measurement methods changed. Bring the source documents—not only the AI summary—to the appointment.

Urgent symptoms require immediate professional help through the appropriate local service. For non-urgent questions, the best output is often a concise appointment brief: verified facts, unresolved discrepancies, questions for the clinician, and decisions that remain with the patient and care team.