JAMA Network Open Survey Shows 63% of Patients Reject Unreviewed AI Responses for Medical Queries

The cross-sectional study used vignettes to measure stated preferences for response speed versus oversight. For logistical tasks such as refills, 57% accepted immediate AI output; for clinical questions, two-thirds chose a 48-hour wait for human editing. Mental-health support showed steeper resistance: 64% preferred a licensed therapist despite a three-month delay, versus 44% for an immediate AI-only chatbot. Demographic stratification revealed that respondents reporting medical discrimination were significantly more open to AI for diagnostic queries, suggesting distrust of traditional systems may lower the threshold for algorithmic alternatives.

Prior observational work, including a 2023 NEJM AI study on clinician-AI collaboration and a 2024 BMJ Quality & Safety analysis of portal message accuracy, indicates that unedited large-language-model outputs carry measurable error rates for nuanced symptoms. The present findings align with those data by showing patients implicitly price that risk when they elect delayed human review. Implementation models that default to AI-first pathways may therefore encounter uptake friction and widen disparities if oversight options are not transparently offered.

Next steps require longitudinal trials linking these preferences to actual engagement metrics and clinical outcomes rather than hypothetical scenarios. Regulatory guidance on disclosure and opt-out mechanisms should be tested against the observed heterogeneity by race, health literacy, and discrimination history.