Study shows AI chatbots give inaccurate medical information half of the time

Artificial intelligence chatbots give inaccurate and incomplete medical information in response to medically related questions about 50% of the time, a British Medical Journal Open study concludes. Thirty percent of responses were “somewhat problematic” and 19.6% were “highly problematic.” The researchers audited Gemini (Google), DeepSeek (High-Flyer), Meta AI (Meta), ChatGPT (OpenAI) and Grok (xAI), asking 10 open- and closed-ended questions about cancer, vaccines, stem cells, nutrition and athletic performance and requesting scientific references. The researchers designed the questions to mimic common information-seeking health questions and information themes found online and in academic discussions and to direct models toward misinformation or advice differing from medical standards. They say continued deployment of AI chatbots without public education and oversight could lead to heightened misinformation. (CIDRAP news article, 4/15/26)