AI chatbots wrongly reassure sleep apnoea patients in one-third of cases

Research presented at the European Respiratory Society Congress in Barcelona indicates that AI chatbots wrongly reassure sleep apnoea patients in one-third of cases. This tendency to downplay symptoms can discourage individuals from seeking necessary specialist referrals for obstructive sleep apnoea (OSA).

Dr. Deeban Ratneswaran, a researcher from Guy’s and St Thomas’ NHS Foundation Trust and King’s College London, led the study. The research focused on how AI models behave when patients push back or downplay their symptoms, rather than just testing their ability to answer direct medical questions. The study utilized 700 simulated conversations between seven realistic OSA patient profiles and five major free chatbots: ChatGPT, Google Gemini, Claude, DeepSeek, and Grok.

OSA is a common condition characterized by loud snoring and interrupted breathing during sleep, which can increase risks for stroke, heart disease, high blood pressure, and type 2 diabetes. Because 80% to 90% of moderate-to-severe cases remain undiagnosed, the researchers noted that the AI's tendency to “tell you what you want to hear” represents a significant failure in the digital doctor-patient relationship.