You open ChatGPT, type a question about your latest blood test – and instead of vague guessing, the AI answers using your actual lab values, your sleep data, your heart rate. Sounds like the future. For most of us in Europe, it's still just that: the future.
In late July, OpenAI rolled out "Health in ChatGPT" to all users in the US – a dedicated health space that connects to Apple Health and real medical records. If you're sitting in Berlin, Vienna, or Zurich, you're currently out of luck.
What ChatGPT Health actually does
US users aged 18 and up can link their Apple Health data and medical records to ChatGPT. The result: a dashboard showing lab values, medications, activity, and sleep – plus the ability to ask questions about any of it in any chat, not just in a separate health tab. Handy if you want to know whether your cholesterol is in the healthy range without digging up old lab reports.
The feature currently runs on every plan, including the free tier. The difference lies in the model behind it: paying users get the current flagship model, while the free version gets a weaker one – according to OpenAI's own figures, answer quality on the medical benchmark "HealthBench" differs noticeably as a result. In short: free works, but with a real trade-off in reliability.
Why Europe is still on the sidelines
OpenAI has said next to nothing about the EU, the UK, or Switzerland – aside from a waitlist that users there can join too. The company hasn't given much of an official reason, but the likely one is fairly obvious: under the GDPR, health data counts as a specially protected category, and the requirements around consent, storage, and purpose limitation are considerably stricter in Europe than in the US. A feature that links medical records to a chatbot has to clear those hurdles first before it can go live here.
This isn't an isolated case, either. New AI features have settled into a familiar pattern by now: launch in the US first, then arrive in Europe later – sometimes months later. If you want a sense of how uneven that rollout pattern generally is, our privacy comparison of the big AI providers is a good starting point.
Privacy: what OpenAI promises
In fairness: OpenAI is promising more restraint here than usual. Connected health data, the company says, won't be used to train its models or for advertising – regardless of whatever training setting is otherwise active on your account. On top of that, this data is supposed to carry stronger encryption than regular chats.
Whether and how that's verifiable in practice remains an open question – true of every "we promise not to use your data" claim, no matter which company makes it. Still, it's a different tone than the usual "everything gets trained on unless you actively opt out" – and that alone isn't a bad signal.
What you can do right now
Right now, trying the feature directly requires a US account and a US location. Routing around that with a VPN violates the terms of service and fails on more than just the geo-block anyway: the actual connection to real medical records relies on US healthcare systems that simply don't exist in Europe – no VPN on earth conjures up a US doctor's office with a matching data export. At most, Apple Health data (sleep, heart rate, activity) could be smuggled through with some trickery, but the actual medical records stay out of reach. Realistically, your options are: wait, or join the waitlist to stay informed once it reaches the EU.
Until then, the same rule applies as with any AI tool and sensitive data: read the privacy policy first, connect your data second – not the other way around. Health data isn't just any data. A hallucinated answer about a prescription sounds funnier than it actually is.
To be fair, though: doctors make mistakes too – especially when several practices prescribe independently of each other and interactions between medications get missed. Whether specialized AI systems will end up making fewer errors than humans in the long run is an open question. But the debate echoes self-driving cars: they're not error-free either – just, on balance, less error-prone than human drivers.