OpenAI has launched ChatGPT Health in the United States, allowing users to connect their medical records, Apple Health data, and wellness apps directly to ChatGPT. The idea is straightforward: you give the tool access to your health information, and it helps you understand lab results, prepare questions before a doctor's visit, or make sense of patterns in your sleep and activity data.
The feature has been in testing since January 2026. OpenAI says health questions are already one of the most common reasons people open ChatGPT, with over 300 million people a week sending health-related messages. That number alone tells you this was always going to happen. OpenAI is not creating a new behaviour; it is building a dedicated product around one that already exists at enormous scale.
Here is where the two-tier structure becomes a real concern. Free users receive health responses from GPT-5.5 Instant, while paying subscribers get access to GPT-5.6 Sol, the newer and more capable model. A subscription starts at $20 a month. OpenAI will likely point out that both models perform well on standardised health tests, but those tests happen under controlled conditions, without the complexity of real patients and real symptoms.
Real-world performance tells a more complicated story. A study published in Nature Medicine in early 2026 tested ChatGPT Health across nearly a thousand medical scenarios. When situations were genuine emergencies, the chatbot failed to recommend urgent care in roughly half the cases. When situations were low-risk, it over-recommended doctor visits nearly two-thirds of the time. The tool performs reasonably well in the middle, on common and moderate-risk situations, which is also where the stakes are lower.
The overconfidence problem makes this worse. Separate radiology research found that AI systems gave incorrect findings with high confidence rather than saying they were unsure, while human specialists were far more willing to admit uncertainty. A tool that sounds certain when it is wrong is more dangerous than one that hedges.
That said, the case for using it carefully is real. Studies show AI can flag patterns in health data that clinicians miss. Tools like AMIE, a Google system tested in simulated consultations, have performed comparably to primary care doctors in structured settings. And OpenAI's own data shows most health questions come in outside clinic hours, when the alternative is either anxious waiting or an emergency room visit for something that did not require one.
For anyone outside the United States, this feature is not available and there is no confirmed timeline. OpenAI explicitly excluded the European Economic Area, Switzerland, and the United Kingdom at launch. The EU's data privacy rules treat health data as a special category requiring explicit legal grounds to process. A recent GovAI study found that EU data protection rules are now causing delays or blocks for around 11 percent of advanced AI feature releases compared to the United States, and health is exactly the kind of area where that friction is highest.
The clearest practical takeaway is this: ChatGPT Health is a reasonable tool for understanding medical language, preparing questions, and getting a first read on health information outside office hours. It is not a triage system, and it should not be the last word on anything urgent. If you use it and pay nothing, you are getting a meaningfully different product than the person next to you who pays $20 a month. In health contexts specifically, that is worth knowing before you trust the answer.