19 October 2026 · Medows
950,000 Clinicians Use OpenEvidence. Now What?
A JAMA study of 950,000 US clinicians shows AI is already in clinical decisions. It does not yet show it helps patients.
A resident on night call has a patient with a problem outside her comfort zone. Nobody senior is free. She opens a tool on her phone, types the question, and gets an answer with citations in seconds. Nobody logs it. Nobody knows it happened.
Multiply that by a few hundred thousand. That is what a new JAMA study, published on October 15, 2026, has now measured.
What the study found
Researchers at Harvard Medical School and Brigham and Women's Hospital analyzed use of OpenEvidence, an AI clinical information platform, by about 950,000 US physicians, physician assistants and nurse practitioners over roughly two and a half years.
The numbers, as reported by Harvard Medical School:
- Monthly active users grew from 355 in January 2024 to 703,265 in June 2026.
- Those practitioners submitted 270 million queries.
- Researchers had a large language model classify a sample of 38.8 million of them.
- Clinical decision-making, particularly treatment and management, made up 51.9% of queries. Diagnosis made up 13.9%.
- Internal medicine physicians generated 27.3% of physician queries, against 14.5% of the physician workforce.
- Clinicians often asked about topics outside their own specialty. A surgeon asking a non-surgical question is the example the authors give.
Anupam Jena, the corresponding author, said an "enormous number of physicians are using AI for clinical care," often where they lack specific expertise. He also noted that adoption was much faster than with electronic health records, which only really took off after government regulation.
What the study does not say
This matters, and the authors say so themselves.
The study does not show whether AI use improves care. That is the next step. It covers one platform only. Query content was classified by a language model on a random sample, not on every query. And three co-authors are OpenEvidence employees. The company did not fund the work but supplied the de-identified data. The research itself was funded by the NIH.
None of that makes the finding wrong. It means the finding is about behavior, not outcomes. Doctors are using these tools heavily. Whether patients are better off is still open.
Why the out-of-specialty pattern is the interesting part
Most coverage will focus on the headline user count. The more useful detail is where the questions come from.
A tool used mostly inside a clinician's own specialty is a faster textbook. A tool used outside it is something else: a stand-in for the colleague you would have asked, if they had been there. On a ward, that colleague is often not there. The registrar is in theatre. The consultant is at home. The question still has to be answered before the next patient.
That is a real need, and the data suggests clinicians are filling it on their own, with a general reference tool, one query at a time.
The gap between an answer and a decision
Here is the founder's view, and it is an opinion, not a finding from the paper.
A reference answer is not a clinical decision. The decision depends on this patient: the creatinine from this morning, the drug chart, the note from last night. A tool that answers the textbook question without seeing the chart leaves the doctor to do the joining in their head, at the end of a long shift.
That is the case for a workspace over a point tool. The answer should arrive next to the patient's own data, with its sources visible, so the doctor can check it rather than trust it. Jena's own caution points the same way: the field needs study of "accuracy and effects on patient care."
What to watch next
Three questions follow from this paper.
- Does heavy use of AI reference tools change outcomes, in either direction?
- Do early-career clinicians, who the study found submitted a higher share of queries, build skill or lean on the tool?
- Who audits a tool that hundreds of thousands of clinicians now use without any hospital procurement process?
Usage has come first, as it usually does. Evidence will have to catch up.
Sources
- Harvard Medical School, "First Large-Scale Study Shows How Doctors Use AI in Medicine" (JAMA study, Patel et al., October 15, 2026): https://hms.harvard.edu/news/first-large-scale-study-shows-how-doctors-use-ai-medicine
Medows is a clinical AI workspace for the doctor on rounds. Learn more or write to alapan@medows.ai / alapanx@gmail.com.