Artificial intelligence chatbots may not be as reliable for medical advice as many users believe, according to new research that warns of potential health risks.
Despite their ability to perform well in medical exams, AI chatbots often fail to provide accurate health guidance to real users, a study published on Monday has found.
“Despite all the hype, AI is not ready to replace a physician,” said study co-author Rebecca Payne of Oxford University.
The researchers cautioned that relying on large language models to assess symptoms can be dangerous, as chatbots may give incorrect diagnoses or fail to recognise when urgent medical attention is required.
The British-led study examined how effectively people could identify health problems and decide whether to seek medical care using AI chatbots, compared with traditional internet searches.
Nearly 1,300 participants in the UK were presented with 10 common health scenarios, including headaches after drinking alcohol, exhaustion in new mothers, and symptoms of gallstones. Participants were randomly assigned to use one of three chatbots — OpenAI’s GPT-4o, Meta’s Llama 3, or Command R+ — while a control group used standard search engines.
The results showed that chatbot users correctly identified their condition only about one-third of the time, while just 45 percent chose the appropriate next step, such as seeing a doctor or going to hospital. These outcomes were no better than those of participants using internet searches.
The findings, published in the journal Nature Medicine, highlight a gap between AI performance in controlled medical tests and real-world use.
Researchers attributed this gap to communication challenges, noting that users often failed to provide complete information or misunderstood the advice given by chatbots.
The study comes amid growing use of AI for health-related queries, with researchers noting that one in six US adults now consults chatbots for medical information at least once a month — a figure expected to rise.
“This is an important study because it shows the real medical risks posed by chatbots,” said David Shaw, a bioethicist at Maastricht University, who was not involved in the research. He advised the public to rely on trusted medical sources such as national health services and qualified professionals.
