Study: AI Meant to Save Doctor Time Is Actually Draining It
New research from Dartmouth College reveals that correcting AI's medical mistakes and missing follow-ups is taking physicians more time than it was intended to save.
A new study from Dartmouth College indicates that the adoption of artificial intelligence (AI) in healthcare may not be speeding up physician workflows as anticipated. Instead, correcting AI-generated responses for errors or inaccuracies, and adding missing follow-up questions, is consuming valuable physician time that could be dedicated to patient care.
The research, presented at the 2026 Annual Meeting of the Association of Computational Linguistics, analyzed over 146,000 patient-physician conversations within the Dartmouth Health system. This represented a large-scale evaluation of online patient portals that utilize AI to draft responses.
Researchers found that AI-generated responses frequently misaligned with what physicians would typically write. Issues included responses that were too lengthy, contained irrelevant or incorrect medical details, or lacked necessary follow-up inquiries. Consequently, physicians must invest time in rectifying these AI-generated outputs.
"We find that AI can sound like a doctor but not think like one," stated Sarah Preum, the study's corresponding author. The findings suggest a need for improved AI training methods in healthcare applications to better support clinicians and enhance efficiency.