Your Artificial Intelligence (AI) assistant wants to replace your GP.
OpenAI — the parent company behind ChatGPT — has launched ChatGPT Health in the United States, letting users import medical records and health-tracking data directly to the clever chatbot.
The AI company, which has a valuation between $850 billion and $1.16 trillion, says it launched the new portal due to the sheer volume of users who already flood ChatGPT with health-related queries. According to OpenAI, some 300 million ask the 'bot about ailments every single week — a dramatic increase from the 230 million when the tool was first tested back in January. ChatGPT users can now pull in data from services like Apple Health, MyFitnessPal, and Function, as well as medical records from hospital systems, including Epic and Oracle Health.
Previously, you had to visit a dedicated health tab to ask these kinds of questions. But OpenAI found that 70% of health queries were actually happening in the main chat, so now the feature works everywhere in the app.
During a press briefing, OpenAI's vice president of health product, Ashley Alexander, said the company's models "are now capable of reasoning at levels that are better than clinician level."
But when pressed on it, OpenAI's health lead Karan Singhal dialled it back a touch, saying he would "temper" that claim. He pointed to "individual studies that have been pointing in that direction," including research from Harvard and Stanford.
A study published in Science earlier this year found that OpenAI's reasoning model o1 could diagnose real-life medical cases as well as — or better than — doctors.
OpenAI also said its newest and smallest model, GPT 5.6-Luna, beats its predecessor on HealthBench, the company's own open-source benchmark for testing how well AI handles health questions.
However, this still might not be the safest method for seeking medical advice.
While AI can provide helpful general information, it is not a replacement for a qualified healthcare professional. The biggest danger is that AI can sometimes give answers that sound confident and accurate but may be incomplete, outdated, or wrong.
This is because AI does not actually understand your health in the same way a doctor does. It generates responses based on patterns in the information it has been trained on and the details you provide, rather than carrying out a medical assessment. If key information is missing, such as your medical history, current medications, lifestyle factors, or the severity and duration of your symptoms, it may produce advice that does not fully apply to your situation.
For example, if someone describes symptoms like fatigue, dizziness, or stomach pain, AI may suggest several possible causes, but it cannot physically examine the person, order tests, or recognise subtle signs that a healthcare professional might notice.
Just one day before the announcement, a 55-year-old Florida pastor named Scott Winters filed a lawsuit against OpenAI and its Chief Executive Officer Sam Altman. Winters claims that an older version of ChatGPT — the now-retired model 4o — gave him "personalised and inaccurate" medical advice that nearly killed him.
He'd started asking the chatbot about his health back in 2024, after feeling dismissed by doctors following a string of health problems. Over time, ChatGPT essentially stepped into the role of a medical practitioner, offering specific care instructions and even prescription drug protocols — without telling him to see a doctor.
When Mr Winters reported groin tenderness, a symptom linked to his developing pulmonary embolism, the chatbot largely brushed it off. It even recommended he limit movement — which one of his physicians later said likely worsened the blood clots in both his lungs.
Ms Alexander responded to this in a statement to Mashable: "We think that treating chatbots as the whole story behind people's medical decisions just really oversimplifies everything here.
"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 also stressed that OpenAI is "very clear" that ChatGPT shouldn't be used for diagnosis or treatment — something the company's own terms of service spell out.
OpenAI pointed to those same clauses when responding to the lawsuit, and told the New York Times it's working to make health-related answers safer. The company also says it doesn't use any personal health data to train its models.
Doctors who've studied how AI handles health questions say people need to be careful about how they use these tools.
Researchers Dr Ashwin Ramaswamy and Dr Girish N. Nadkarni from Mount Sinai Health System found that when users included dismissive phrases like "I'm probably overreacting" or "My wife thinks it's fine," ChatGPT Health was 11 times more likely not to send them to A&E — even when their symptoms pointed to something life-threatening, per a peer-reviewed paper in Nature Medicine.
Their upcoming research also shows that long, ongoing chats tend to drift away from proper triage standards. Their advice? Start a fresh conversation for every new health concern.
They also recommend pushing back on the chatbot's answers to see if it holds firm or simply agrees with you.
Both called on OpenAI to open its products to independent testing. "If the systems are as good as the statements claim, independent measurement is the fastest way to prove it," they said.
The new feature works across free and paid plans on both the web and iOS in the US only.
