OpenAI is turning ChatGPT into something much closer to a health companion, and it wants your doctor’s notes, lab results, and step counts to do it. The new “ChatGPT Health” tab invites people to plug in their medical records and data from wellness apps so the chatbot can talk to them about their bodies with far more context than a generic web search ever could.​

On the surface, the pitch is simple: health is already one of the biggest use cases for ChatGPT, with OpenAI saying more than 230 million people a week ask it about symptoms, lab tests, diets, and exercise. Instead of tossing those questions into the same general-purpose chat window you use for emails and math homework, Health lives in its own sandboxed space, with separate memory and its own chat history. OpenAI is rolling it out gradually via a waitlist, but it’s not locking it behind a paid tier, which signals that the company sees this as a mainstream feature, not a premium add-on for power users.​

What makes Health different is the level of intimacy OpenAI is asking for. The company is actively encouraging people to connect their patient portals through a partner called b.well, which plugs into a network of roughly 2.2 million providers in the US. Once connected, ChatGPT can see things like lab results, visit summaries, and bits of clinical history, then blend that with data from apps like Apple Health, MyFitnessPal, WeightWatchers, Peloton, and Function. In theory, that lets it go beyond generic “eat less processed food” advice and talk to you about your cholesterol trend line, your sleep patterns, and what actually shows up in your lab work.​

OpenAI is careful to stress that this isn’t a diagnostic engine and emphatically says Health is “not intended for diagnosis or treatment,” framing it instead as a way to help people prepare for appointments and understand the trade-offs between treatments or insurance options. The company points out that seven in 10 health conversations in ChatGPT already happen outside normal clinic hours, and that usage is especially heavy in rural and underserved communities, where professional care can be hard to reach. The subtext is clear: if people are going to ask a chatbot about their health anyway, OpenAI would rather give them a purpose-built space with tighter controls than leave them in the free-for-all of the main app.​

Behind the scenes, OpenAI is leaning heavily on clinician optics to make this feel legitimate. It says more than 260 physicians across dozens of countries have provided feedback on model outputs over the past two years, reviewing answers more than 600,000 times across around 30 areas of focus. That testing is packaged as part of a broader “OpenAI for Healthcare” push, which also includes GPT-5–era models tuned for clinical workflows and APIs pitched at hospitals, insurers, and health-tech startups. The message to the medical industry is: your patients are already here — let’s give you tools to meet them where they are.​

Still, the company is walking into a minefield. There is a very real history of chatbots giving dangerous or simply bizarre medical advice, and not just from small players. Google’s AI Overview famously suggested adding glue to pizza and has been caught surfacing misleading or harmful guidance on cancer screenings and lab tests, and doctors have documented cases where people followed AI suggestions that led to serious harm, including hospitalization. OpenAI itself has been cited in case reports where people described relying on ChatGPT for medical guidance, sometimes with devastating consequences, which is why that “not for diagnosis” disclaimer is doing so much work here.​

Mental health is the most conspicuous gap in OpenAI’s official messaging. In the launch blog post, the company largely sidestepped explicit promises around therapy-like support, opting for a vague line about letting users customize instructions “to avoid mentioning sensitive topics.” Yet during a briefing, OpenAI’s head of applications acknowledged what everyone already knows: a lot of people already talk to ChatGPT about anxiety, depression, and other mental health struggles, and Health will also handle those conversations. OpenAI says it is focusing on routing people in distress toward professionals, loved ones, or crisis resources, but there is still no way to fully control what people do with the emotional and medical advice they receive at 2 am from a bot that never sleeps.​

There’s also the quieter but equally important risk of amplifying health anxiety. If you tend to doomscroll your symptoms on search engines, it’s easy to imagine spiraling even faster when you have a persistent, context-aware assistant that remembers your past worries and lab quirks. OpenAI says it has tuned the model to be “informative without ever being alarmist,” and to redirect users to the healthcare system when action is needed, but this is a very thin line to walk when the model is built on probabilities, not clinical judgment. For people with hypochondria or obsessive health worries, a 24/7 chatbot that knows every blip in their blood work could become a new kind of trigger.​

Privacy is where the stakes feel highest. OpenAI is effectively asking people to pipe some of the most sensitive data they possess into a system that, until now, has been synonymous with consumer-grade AI experimentation. The company says Health runs in a separate space with “enhanced privacy,” uses multiple layers of purpose-built encryption, and keeps Health conversations and memories out of its foundation-model training by default. But it is not end-to-end encrypted, and OpenAI has already had a notable security incident in 2023, when a bug briefly exposed some users’ chat titles and account details to others.​

Legally, the situation is nuanced in a way most consumers will never see. OpenAI’s head of health has said that HIPAA — the US health privacy law — doesn’t apply to ChatGPT Health in the same way it does to hospitals or clinics, because this is a consumer product, not a covered clinical entity. That means your rights look different depending on whether you’re using a hospital’s enterprise deployment of OpenAI (where HIPAA may apply under a business associate agreement) or casually syncing your personal records to the ChatGPT app on your phone. And OpenAI notes that it can still be compelled to hand over data in response to court orders or emergencies, something that will matter more as health data gets tangled up with everything from insurance disputes to criminal investigations.​

For now, OpenAI is trying to split the difference between utility and restraint. Health will nudge you to move sensitive conversations into its dedicated space, where the company says the privacy rules are stricter, but it’s also trying to keep the experience casual enough that you’ll actually use it. You can ask about insurance trade-offs, get a plain-English summary of your latest blood panel, or brainstorm questions to bring to your next appointment, all in the same chat where you might also store a grocery list or a workout plan. That kind of convenience is exactly why people will try it — and why privacy advocates are deeply uneasy about what happens once health data becomes just another input to the world’s most popular chatbot.​

Zoom out, and ChatGPT Health is part of a much bigger shift: AI models moving from generic assistants into domain-specific infrastructure that sits under real healthcare workflows. Hospitals are already piloting OpenAI’s models for things like automated clinical documentation, discharge summaries, and triage notes, often via HIPAA-compliant API setups. ChatGPT Health is the consumer-facing tip of that spear — a way to get patients accustomed to the idea that an AI system might be reading their charts, summarizing their doctor visits, and nudging them about follow-ups.​

Whether that future feels empowering or dystopian will depend on how well OpenAI handles the next phase. If Health consistently helps people understand their bodies, make better use of their doctors’ time, and avoid missed red flags, it will be easy to argue that the trade-off was worth it. If, on the other hand, people see targeted ads that feel a little too informed, hear about another breach, or watch a friend spiral after a late-night chat with a model that sounded confident but was quietly wrong, trust could evaporate quickly.​

For now, ChatGPT Health sits in a familiar gray zone: a powerful, polished tool that’s undeniably useful and undeniably risky, wrapped in careful disclaimers and strong but imperfect privacy promises. OpenAI is betting that millions of people will be willing to trade a new level of intimacy for personalized explanations and a feeling of control over their health. The question is whether they fully understand what they’re giving up when they click “connect records.”