AI Gives 'Problematic' Health Advice Around Half The Time, Study Suggests

A BMJ Open study found that about half of answers from five major AI chatbots were judged problematic on health questions.

Performance was especially weak on open-ended prompts, and some responses included inaccurate or fabricated references.

ChatGPT, Gemini, Grok, Meta AI, and DeepSeek all raised reliability concerns in health advice use.

The study warns that chatbots can sound authoritative while giving misleading or unsafe medical guidance.