One in six American adults asks ChatGPT for medical advice at least once a month. Spoiler alert: that’s a terrible idea.
A new study out from Oxford University confirms what your actual doctor has been saying for months now — AI chatbots are shockingly bad at helping real humans figure out real health problems. As in, worse than just Googling your symptoms bad.
“Despite all the hype, AI just isn’t ready to take on the role of the physician,” says Rebecca Payne, who led the Oxford research. “Patients need to be aware that asking a large language model about their symptoms can be dangerous, giving wrong diagnoses and failing to recognize when urgent help is needed.”
Researchers gave nearly 1,300 people realistic health scenarios — hangover headache, new mom exhaustion, gallstone symptoms — then turned them loose with either ChatGPT, Meta’s Llama 3, Command R+ or just regular Google.
The AI users correctly identified their “condition” only about a third of the time. They figured out the right next step (ER? Urgent care? Sleep it off?) less than half the time. Google users? Literally just as good.
Here’s the kicker: these same chatbots ace medical licensing exams. They can pass tests designed for doctors. But put them in conversation with an actual anxious human Googling symptoms at 2 a.m.? Total disaster.
The problem isn’t the AI’s medical knowledge — it’s that humans are messy communicators. We don’t give chatbots all the relevant info. We misunderstand their answers. We straight-up ignore their advice. The simulated patient scenarios that AI crushes in testing don’t translate to real-world panic and incomplete information.
And here’s the thing: we’re putting an awful lot of faith in technology that clearly isn’t ready for it. ChatGPT sounds confident and authoritative, which makes it easy to trust — even when it’s spectacularly wrong. But confidence isn’t the same as competence, and the stakes are real when we’re talking about actual human health.
If AI can’t reliably help you figure out whether your headache needs Advil or an ER visit, it’s worth asking what other major life decisions we’re outsourcing to chatbots. Relationship advice? Parenting questions? Ethical dilemmas? Financial planning? AI might be helpful for brainstorming or organizing information, but treating it like an all-knowing oracle — especially for high-stakes decisions — is a gamble we’re not equipped to make yet.
“This is a very important study as it highlights the real medical risks posed to the public by chatbots,” says David Shaw, a bioethicist at Maastricht University who wasn’t involved in the research.
Translation: asking ChatGPT “is this mole weird?” could actually hurt you. The bot might miss something urgent. Or it might send you spiraling over nothing.
Shaw’s advice? Stick to actual medical sources — like the NHS website if you’re in the U.K., or the Mayo Clinic if you’re stateside. Or, wild idea: call your doctor.
As AI becomes more ubiquitous (and honestly, more persuasive), more people will be tempted to skip the copay and ask their phone. This study is a reminder: ChatGPT might sound confident, but that doesn’t make it qualified. Your health deserves better than a chatbot playing doctor.
