Artificial intelligence chatbots do not consistently help users correctly identify medical conditions or decide when professional care is needed, according to a recent study published in the journal Nature Medicine.
The British-led research examined how people use AI chatbots to assess health concerns and determine appropriate next steps such as self-care, seeing a doctor, or seeking emergency treatment. Nearly 1,300 participants in the United Kingdom were presented with 10 common medical scenarios, including headaches after alcohol consumption, fatigue in new mothers, and symptoms linked to gallstones.
Participants were randomly assigned to consult one of three AI systems — OpenAI’s GPT-4o, Meta’s Llama 3, or Command R+ — while a control group relied on standard internet search engines. The study found that participants using chatbots correctly identified the medical issue in about one-third of cases. About 45 percent selected the appropriate course of action. These results were similar to those of participants using search engines.
The researchers noted that the findings contrast with the high scores AI systems achieve on medical exams and benchmarks. They attributed the difference to how people interact with chatbots in real-world settings, where users may provide incomplete information, misunderstand responses, or fail to follow guidance given by the system.
Study co-author Rebecca Payne of University of Oxford said the results show limitations in how large language models currently function when used by the public for health-related questions.
The study also cited data indicating that around one in six adults in the United States consult AI chatbots for health information at least once a month, a figure expected to rise as adoption increases.
David Shaw, a bioethicist at Maastricht University who was not involved in the research, said the findings highlight risks associated with using chatbots for medical guidance and emphasized reliance on established healthcare providers and official medical institutions for health information.
