What the research found
Harvard Medical School and Brigham and Women's Hospital researchers analysed de-identified usage logs from OpenEvidence, an AI clinical information platform, covering about 950,000 US physicians, physician assistants and nurse practitioners from January 2024 to June 2026. Monthly active users rose from 355 to more than 703,000, and in a random sample of 38.8 million queries classified by a language model, 51.9% concerned clinical decision-making and 13.9% diagnosis, often outside the asker's own specialty. The study measured use, not accuracy or patient outcomes: the authors name assessing accuracy and whether these tools improve care as the next step, and the platform supplied the data, with three co-authors employed by it during the study.
AI is already shaping bedside decisions at national scale, so providers need their own view of when clinicians use these tools and how answers get checked, rather than waiting for outcome evidence to catch up.