OpenAI flipped the switch on ChatGPT Health this week, opening it to any US user over 18 and inviting them to plug in their medical records alongside data from Apple Health, Function, and MyFitnessPal. The consumer hook is obvious: a chatbot that already answers your health questions can answer them better if it knows your labs, your steps, and your history. But the framing OpenAI chose is what makes this launch worth pausing on.
The claim doing the heavy lifting
During the briefing, OpenAI’s VP of health product Ashley Alexander said the company’s models “are now capable of reasoning at levels that are better than clinician level.” That is a remarkable sentence, and it is doing an enormous amount of work. “Reasoning at clinician level” on a benchmark is not the same as practicing medicine. Doctors don’t lose to a model because they can’t pattern-match symptoms; they carry liability, examine patients, order tests, and get sued when they’re wrong. A benchmark score, however high, doesn’t inherit any of that accountability. When a company markets a superlative like this while simultaneously asking you to hand over your health records, the claim and the ask should be read together, not separately.
The real product is the data pipe
Strip away the messaging and ChatGPT Health is, functionally, a consented ingestion point for some of the most sensitive data a person owns. The integrations with Apple Health, Function, and MyFitnessPal matter more than the model boast, because they determine what OpenAI can actually see and retain. This is the second time OpenAI has made this pitch — Engadget’s framing of “once again” is telling — and the repetition suggests the harder sell isn’t capability, it’s trust. General-purpose consumer AI isn’t covered by HIPAA the way a hospital’s systems are, and “connect your records” is a one-way door most users won’t think hard about.
Why the framing should make you cautious
There’s a pattern in AI product launches: lead with a benchmark-flavored superlative, bury the disclaimers. “Better than clinician level” is the headline; “this is not medical advice, consult a professional” is the footnote. For a low-stakes question — is this rash worth a call? — the tool is genuinely useful, and millions will use it that way. The risk is the medium-stakes case, where a confident, personalized, record-aware answer nudges someone away from a doctor they should have seen.
ChatGPT Health will be useful. The technology is real and the convenience is real. But “better than clinician level” is a marketing line, not a clinical standard — and the price of admission is your medical history. That’s a trade worth making slowly, and with eyes open.