OpenAI on Thursday made its Health in ChatGPT feature broadly available, enabling U.S. users to connect its artificial intelligence chatbot with medical records and wellness apps.
With the medical records integration, the AI company aims to shift ChatGPT from a general health Q&A tool into a more personalized health companion that can draw on users’ medical and wearable data for deeper context.
Every week, more than 300 million people turn to ChatGPT with health-related questions, according to OpenAI, and that number has grown from 230 million in January.
The Health in ChatGPT feature, initially announced in January, is designed to help users understand and navigate their health information, OpenAI executives said during a media briefing, by drawing together information that's often scattered in patient portals, medical records, apps and wearables.
ChatGPT is not intended to be used for diagnosis and treatment and is not designed to replace the care and judgment of qualified medical professionals, OpenAI executives emphasized.
"The reality is, we're living in a time when not everyone has access to quality care or care in a timely manner. The average doctor appointment in the United States is less than 15 minutes. You're often left on your own to sort through a lot of fragmented information to make sense of it all to try to navigate next steps," Ashley Alexander, VP of Health Products at OpenAI, said during a briefing with reporters. "Chat has been really helpful for people in preparing for those doctor appointments, preparing for next steps, digesting information and building healthier day-to-day routines."
Users can connect Apple Health data and supported medical records, including records available through hospital systems connected to Epic and Oracle Health electronic health record systems, along with One Medical and Function Health, OpenAI executives said.
"We are starting with the sources that provide the clearest and most broadly useful health context and will continue to explore additional integrations," an OpenAI spokesperson said. "If an app integrates with Apple Health, ChatGPT can use the data it makes available there, with permission. Users can also combine connected health context with other apps and connectors in a conversation, subject to additional safeguards for sensitive actions."
OpenAI announced the Health in ChatGPT feature in January and tested it with early waitlist testers. The company used feedback from those users to refine the experience while also continuing to improve the models that power it, Alexander said.
OpenAI also collaborates with a global network of hundreds of physician advisors across 60 countries to train and evaluate its medical intelligence features like ChatGPT Health, executives said. These physicians test the model's performance on health questions and they develop realistic health scenarios and detailed rubrics that define what a strong response should do, according to OpenAI in a blog post. These evaluations assess qualities including accuracy, safety, communication, context awareness, completeness, and appropriate escalation.
"A model can't just be able to answer textbook health or medical questions well. They have to be able to answer real-world questions from real users, and to do this in a way that makes sense, we have to test situations that are ambiguous and messy and noisy and underspecified," Rebecca Soskin Hicks, M.D, a pediatrician and a physician network lead at OpenAI, said during the media briefing. "And now we are evaluating the model's capability to do this with connected information, so not just a user asking a health question, but a user asking a health question who also has connected health data, either through their electronic health record or other connectors."
She added, "We are crossing a rung of our evidence generation ladder that we are continuing to improve on and continue to collaborate closely with our physician community, who really do define what good looks like."
OpenAI executives also touted continued performance improvements of the models. GPT‑5.5 Instant, available to users on the ChatGPT free plan, has made "meaningful gains" in recognizing when urgent care may be needed, asking for relevant context, explaining uncertainty and making complex information easier to understand, executives said. "On our most challenging health evaluations, it performed at a level comparable to our frontier Thinking models at the time of evaluation," the company said in the blog post.
OpenAI said GPT‑5.6 Sol, available to paid users, is the "strongest model yet for health." The model demonstrates stronger performance on tasks that require reasoning across complex details, clear communication and careful judgment. "It can help explain a visit note in plain language, understand how lab results have changed over time, consider the factors across options a doctor discussed, or prepare questions for a follow-up appointment," OpenAI said.
Every GPT‑5.6 model outperformed GPT‑5.5 on HealthBench Professional (opens PDF), an evaluation of AI performance on challenging health tasks, OpenAI claims.
OpenAI executives also said Health in ChatGPT is built with layered privacy and security safeguards. All ChatGPT conversations are encrypted at rest and in transit, and information connected in Health receives additional encryption protections.
The information connected in Health is not used to train OpenAI foundation models or target ads. Connected medical records and Apple Health information, and conversations that use it, are not used to train foundation models or target ads, regardless of the model-training setting the user chooses for ChatGPT, executives said in the blog post. Conversations that do not use this data follow standard ChatGPT model-training setting choices, which users can change at any time, the company said.
Users can decide when ChatGPT can use the information connected in Health. By default, ChatGPT asks for permission before using connected medical records and Apple Health information to personalize a response, according to the company. And users can disconnect their health information at any time.
When OpenAI initially launched Health in ChatGPT in January for early testing, it introduced a dedicated health experience that required visiting a separate space in ChatGPT to get the full benefit.
"What we heard from users was that this really wasn't working," a member of OpenAI's health product team said during the media briefing. "We saw that more than 70% of them were still having health conversations in the main chat experience and when we dug into that, we found it really makes sense. It's really hard for people to separate their health from the rest of their life."
Moving to a separate space added an unnecessary step for users. Based on that feedback, OpenAI designed Health in ChatGPT so users can choose to let ChatGPT draw on relevant information that they have connected in Health across their conversations. With permission, ChatGPT can consider relevant information the user has connected, such as medications, lab results, recent visits, sleep and activity, alongside the goals and context they've shared.
Users control where they can ask health questions, what health data is connected and when that data is used for context, OpenAI executives said.
Millions of people now use AI chatbots to check health symptoms, research conditions or understand medical treatments and bills. But studies, including findings published in Nature Medicine and the BMJ, warn that chatbots can sometimes provide inaccurate, inconsistent, or even dangerous medical advice.
The New York Times reported this week that a man in Florida has sued OpenAI, claiming that its chatbot, ChatGPT, had offered him “extremely dangerous medical recommendations” that led to delayed care for a life-threatening pulmonary embolism last year. The plaintiff in the lawsuit and Tech Justice Law, the nonprofit helping to represent him, are seeking financial damages and asked the court to stop ChatGPT Health from operating until independent evaluators determine it is safe, the NYT reported.
In response to a question from a reporter about the lawsuit, Alexander said, "It's important that we zoom out and ground it in the bigger picture, which is that every day already hundreds of millions of people are already going to the internet seeking information to support them in their health, and AI can really make that experience better by helping those answers that they're finding get clearer for them. And again, in the bigger picture, we're living in a world where not everyone has equal access to quality care or timely quality care," she said. "We think that treating chatbots as the whole story behind people's medical decisions just really oversimplifies everything here, and it really risks actually getting in the way of us getting the beneficial impact of this in the hands of as many people as possible in ways that can really aid them in their health journey."
She emphasized that ChatGPT is not intended for diagnosis or treatment, highlighting extensive physician-led testing and safety evaluations.
"Our latest GPT-5 models are correctly recommending people go to immediate emergency care greater than 99% of the time when it should, and we also are seeing, against those same evaluations, that it's also importantly avoiding unnecessary escalation to emergency care also 99% of the time," Alexander said.