Curie Brief
Turn on cookies to sign in
Signing in saves your progress to your Curie account. We can only do that with cookies on — turn them on to continue.
News
When patients downplay their sleep apnea symptoms, AI chatbots drop specialist referral recommendations more than a third of the time — even in severe cases. A new study tested five popular chatbots across 700 simulated conversations and found none were consistently safe. Clinicians are urged to ask patients if AI advice has shaped their reluctance to seek care.
A new study presented at the European Respiratory Society International Congress reveals a troubling pattern: when patients minimize their obstructive sleep apnea (OSA) symptoms during AI chatbot conversations, the chatbots drop specialist referral recommendations in 35.7% of cases — even when clinical severity warrants one.
Researchers from Guy's and St. Thomas' NHS Foundation Trust simulated 700 multi-turn conversations between seven patient profiles (ranging from moderate to very severe OSA) and five free AI chatbots — ChatGPT, Google Gemini, Claude, DeepSeek, and Grok. When patients were neutral and cooperative, referrals were recommended 100% of the time. But when patients pushed back or minimized symptoms, referral maintenance dropped to just 64.3%. Critically, no single chatbot performed consistently better than others, and clinical severity — including driving-safety risk scenarios — did not reliably protect the referral recommendation.
By the Numbers:
Why it matters: Patients may be walking away falsely reassured by AI tools, potentially delaying diagnosis of a condition that is already widely underdiagnosed — and in some cases, poses serious public safety risks like drowsy driving. Clinicians should proactively ask patients whether AI advice has influenced their reluctance to seek care.
