Relying on artificial intelligence chatbots for medical guidance could put patients at risk, according to new academic research. The study suggests that while such systems can generate convincing answers, they frequently deliver advice that is inaccurate, inconsistent or potentially unsafe when applied to real-life health concerns.

The investigation was carried out by experts from the Oxford Internet Institute and the University of Oxford’s Nuffield Department of Primary Care Health Sciences. Writing in the journal Nature Medicine, the researchers examined how people used large language model (LLM) tools when presented with a range of medical scenarios. Nearly 1,300 participants were asked to identify possible conditions and decide on an appropriate course of action. Some consulted AI chatbots, while others relied on more traditional routes, including speaking to a GP.

The findings revealed that chatbot responses often blended correct and incorrect information, leaving users unable to reliably tell the difference. Although these systems perform strongly in formal medical knowledge tests, the researchers warned that real-world use is far more complex. In particular, the tools sometimes failed to recognise when symptoms required urgent attention.

Dr Rebecca Payne, a GP and co-author of the study, said the technology was not yet capable of replacing a trained clinician. Lead author Andrew Bean added that human interaction remains a significant hurdle for advanced AI systems, and called for further work to ensure they are safe and genuinely helpful in healthcare settings.