Studies Find ChatGPT Often Misguides Patients on Urgent Medical Care
An NPR report on March 11, 2026 details new research warning that popular AI chatbots like ChatGPT can mislead people seeking medical advice, especially about how urgently they need care. In a Nature Medicine study that tried to mimic how laypeople actually use AI, participants who consulted chatbots correctly identified a hypothetical condition only about a third of the time, and only 43% chose the right next step, such as going to the ER or staying home. A separate study found that in 52% of emergency scenarios, chatbots under‑triaged the problem, including a case of diabetic ketoacidosis with impending respiratory failure where the bot failed to send the patient to an emergency department. Researchers say small differences in wording—such as whether a headache is described as "the worst ever"—can change advice from "go to the ER now" to "take aspirin and stay home," underscoring how non‑experts may not know which symptoms to highlight. OpenAI disputes that the studies reflect typical ChatGPT use and notes they relied on older model versions, but physicians and AI researchers interviewed by NPR argue that while these tools can help explain conditions or prepare for doctor visits, they should not be treated as a substitute for professional triage, particularly in time‑sensitive emergencies.
đź“Ś Key Facts
- Nature Medicine study: after using AI chatbots, participants correctly identified the hypothetical medical condition only about one‑third of the time and chose the correct next step only 43% of the time.
- In a separate triage study, AI chatbots under‑triaged 52% of emergency‑level cases, sometimes failing to direct patients with life‑threatening conditions like diabetic ketoacidosis and impending respiratory failure to the ER.
- Researchers highlight that small changes in how laypeople describe symptoms (e.g., calling a problem "the worst headache I've ever had") can radically change the bot’s recommendations.
- OpenAI says the studies used older versions of ChatGPT and argues they do not mirror how people typically use the tool, while doctors interviewed by NPR say AI can still be useful if treated as a supplement, not a replacement, for clinical care.
đź“Š Relevant Data
About 1 in 8 U.S. adolescents and young adults aged 12 to 21 use AI chatbots for mental health advice, with usage most common among those aged 18 to 21.
One in Eight Adolescents and Young Adults Use AI Chatbots for Mental Health Advice, New Study Finds — RAND Corporation
AI chatbots exhibit racial bias by providing less empathetic responses when detecting Black dialects in user inputs, compared to White dialects.
Study reveals AI chatbots can detect race, but racial bias reduces response empathy — MIT News
Only 30% of Black adults and 36% of Hispanic adults report acceptance of AI in healthcare, compared to higher rates among other groups, indicating disparities in trust and adoption.
Bridging the digital divide: artificial intelligence as a catalyst for health equity — International Journal of Medical Informatics
AI chatbots can perpetuate health disparities through biases in training data, leading to inaccurate or discriminatory medical recommendations for marginalized racial and ethnic groups.
Misuse of AI chatbots tops annual list of health technology hazards — ECRI
đź“° Source Timeline (1)
Follow how coverage of this story developed over time