A study by seven researchers has found that AI chatbots often give "problematic" health advice around half the time, according to BMJ Open. The researchers put five of the world's most popular chatbots, ChatGPT, Gemini, Grok, Meta AI, and DeepSeek, through a systematic health-information stress test. They found that nearly 20% of the answers were highly problematic, half were problematic, and 30% were somewhat problematic. None of the chatbots reliably produced fully accurate reference lists, and only two out of 250 questions were refused to be answered. Grok was the worst performer, with 58% of its responses flagged as problematic. The study also found that while chatbots could get the right medical answer almost 95% of time, they struggled to suggest the right set of possible conditions more than 80%.