“A study published in the medical journal BMJ Open found that 20% of AI chatbot responses to health questions were "highly problematic."”