Most patients won't endorse AI medical responses without a clinician's review, study finds

A Columbia University study of 1,185 U.S. adults reveals that 63% of patients would not approve of AI-generated medical responses without a clinician's review. This finding suggests that while people see potential benefits in AI for certain healthcare tasks, they still favor human oversight when it comes to medical advice. The research, titled "Public Perceptions of Timeliness for AI Use and Human Oversight," was published in the JAMA Network Open on October 14, 2026.

Researchers discovered that patient preferences varied depending on the task. A majority (57%) were more inclined to accept fast AI-generated responses for logistical requests such as scheduling appointments and refilling prescriptions. However, two-thirds of participants were willing to wait up to 48 hours for a doctor or nurse to review and edit an AI response to medical inquiries.

When it came to mental health support, the study found that participants were most likely to opt for a licensed human therapist, even if they had to wait three months (64%). Acceptance for a therapist-monitored AI chatbot available within one month decreased to 50%, and for an AI-only chatbot available immediately, it dropped to 44%.

Natalie Benda, Ph.D., the lead author and assistant professor at Columbia Nursing, emphasized that despite AI's potential to provide immediate answers, patients still prefer speaking with their doctor regarding medical concerns. She noted that people's opinions varied based on factors like race, health literacy, gender, and past experiences of discrimination in medical settings. Interestingly, those who reported medical discrimination were more open to the idea of using AI for medical questions.

The study underscores the significance of enabling patients to decide how and when AI should be incorporated into their medical care. The authors recommend that patients might be more accepting of AI in lower-stakes situations, but they still expect clinician involvement for high-stakes medical decisions and guidance. Other contributors to the research include Uday Suresh, Ph.D., Megan E. Salwei, Ph.D., Laurie Lovett Novak, Ph.D., all from Vanderbilt University Medical Center; Afra Shamnath, MPH, and Meghan Reading Turchioe, Ph.D., from Columbia University School of Nursing; Isabella Buitron, MPH, from Columbia University Mailman School of Public Health; and Xuhai (Orson) Xu, Ph.D., from Columbia University Department of Biomedical Informatics.