ChatGPT is moving from answering general health questions to reasoning over a person's own medical history.
OpenAI has launched Health in ChatGPT for logged-in U.S. users aged 18 and older. People can choose to connect Apple Health and supported medical records, giving the assistant context from lab results, medications, recent visits, sleep, activity, and workouts.
That changes the shape of the conversation. Instead of repeatedly uploading a result or reconstructing a timeline from memory, someone can ask what has changed since an appointment, compare a new test with earlier ones, explore how activity relates to a routine, or prepare sharper questions for a clinician.
OpenAI says more than 300 million people already use ChatGPT for health-related questions every week. Health turns those scattered interactions into something closer to a longitudinal personal reference layer.
What Health can actually do
The practical value is not a dramatic diagnosis button. It is continuity. Medical information usually lives across portals, PDFs, apps, wearables, and half-remembered conversations. Health gives ChatGPT permission to draw on the relevant pieces when answering a question.
A user might ask for a plain-language explanation of a visit note, a summary of lab changes over time, or a list of questions to bring to a follow-up. The assistant can also account for connected context in ordinary conversations—for example, considering a food allergy while discussing restaurants or a recent injury while planning weekend activity.
Health remains the control center for connecting data, browsing synced records, reviewing trends, and managing past health conversations. OpenAI says users can correct information that is incomplete or outdated and decide what is connected and when ChatGPT may use it.
The privacy boundary matters as much as the feature
Medical records are among the most sensitive data a consumer product can handle. OpenAI says connected records and Apple Health information receive additional encryption protections, with access controlled through user permissions.
The company also says connected health information—and conversations that use it—will not be used to train its foundation models or target advertising. Users can disconnect a source and manage the information from the Health experience.
Those safeguards are central, not secondary. A useful health assistant needs enough context to understand a timeline without quietly turning that context into an unrelated data asset. The product will ultimately be judged on whether consent remains legible, revocable, and narrow as the experience expands.
Better evaluation does not eliminate medical risk
OpenAI says it has worked with hundreds of physicians to improve how its models handle health questions. The evaluations test accuracy, safety, communication, context awareness, completeness, and whether the assistant escalates appropriately when professional care may be needed.
HealthBench Professional focuses on difficult clinical tasks such as care consultation, documentation, and medical research. OpenAI also reports improvements in recognizing urgent situations, asking for missing context, explaining uncertainty, and avoiding overconfident answers.
That is evidence of progress, not a guarantee of correctness. A benchmark cannot reproduce every medication interaction, unusual symptom, incomplete record, local care pathway, or emergency. OpenAI explicitly warns that ChatGPT can still make mistakes and should support—not replace—the judgment of qualified medical professionals.
Why it matters
The most important shift is from generic health information to personal health context. A general answer can explain what a lab marker usually means. A connected assistant can notice that the marker changed across three tests, relate it to the rest of the record, and help the user formulate a question for the person responsible for their care.
That could make fragmented records more understandable and help people become better prepared participants in their own healthcare. It also raises the stakes. When an answer feels personal and informed, users may trust it more—even when the underlying interpretation is wrong.
Good product design therefore has to make uncertainty visible, distinguish explanation from medical advice, surface missing context, and create clear escalation paths to clinicians and emergency services.
The SunMarc angle
For SunMarc App Labs, Health in ChatGPT is a strong example of how personalization should be designed around permission rather than silent collection. The value comes from context, but the product earns the right to use that context through clear controls, limited purpose, and honest boundaries.
The same principle applies well beyond medicine. A finance tool, location app, fitness product, or personal assistant becomes more useful when it remembers history and connects data. It also becomes more consequential. The more intimate the context, the more visible consent, deletion, data minimization, and error recovery need to be.
The product lesson
Do not confuse personalization with authority. A product can know more about the user and still be uncertain about the answer.
The strongest version of this experience will not pretend to be a doctor. It will reduce the work of gathering context, translate difficult information, expose trends worth discussing, and help the user arrive at professional care with better questions.
Health in ChatGPT is a meaningful step toward AI that understands a person's history rather than only the latest prompt. Whether that becomes genuinely empowering will depend on the quality of the reasoning, the clarity of its limits, and the durability of the privacy promises around the data.