A Harvard-led analysis of 270 million clinical queries found rapid adoption of one AI platform, with most questions focused on treatment rather than diagnosis.
A Harvard Medical School and Brigham and Women’s Hospital study offers a large-scale look at how clinicians are using AI in practice. Researchers analyzed de-identified activity on OpenEvidence, a clinical information platform, covering approximately 950,000 U.S. physicians, physician assistants and nurse practitioners over two and a half years. The findings were published in JAMA on October 15.
From January 2024 through June 2026, monthly active users in the study grew from 355 to 703,265. Clinicians submitted 270 million queries during that period; researchers used a language model to classify a random sample of 38.8 million. About 51.9% of queries concerned clinical decision-making, especially treatment and management. Diagnosis accounted for 13.9%.
A notable pattern was clinicians asking about areas beyond their specialties. Surgeons, for example, often submitted questions unrelated to surgery. Usage also differed by specialty: oncology and hematology accounted for 3.8% of queries while making up 1.9% of the physician workforce. Internal medicine represented 27.3% of physician queries and 14.5% of the workforce. These figures describe activity on one platform, not AI use across all clinical tools.
The study documents adoption and query types, but does not establish whether AI recommendations are accurate or improve patient outcomes. The researchers say evaluating effects on care is a next step. The platform supplied de-identified data, and three co-authors were OpenEvidence employees; the study was federally funded and the company did not fund it. For health systems, the findings underscore that clinicians are already consulting AI during care-related work, including outside their areas of specialty. The evidence gap now shifts from whether clinicians are using these tools to how safely and effectively they affect decisions and outcomes.