OpenEvidence JAMA study tracks growth to 703,265 monthly users
The OpenEvidence JAMA study, published October 15, 2026 by researchers at Harvard Medical School and Brigham and Women's Hospital, found that monthly active users of the clinical AI platform rose from 355 in January 2024 to 703,265 in June 2026. Across 950,000 US physicians, physician assistants and nurse practitioners, 51.9% of queries concerned clinical decisions. For health systems, the data shows clinicians already use an evidence tool on their own, which is why OpenEvidence is now moving inside the electronic health record.
The question doctors bring to an AI evidence tool
Clinicians often need medical evidence while they treat a patient, including in fields outside their own training. OpenEvidence built an AI-powered medical search engine and a generative AI chatbot exclusively for doctors that summarizes and simplifies evidence-based medical information, according to Fierce Healthcare.
Until this study, there was little hard data on how such tools are used. Anupam Jena, the study's corresponding author, said there has been limited information on how many doctors use AI, what they use it for and how it affects patients. The JAMA paper addresses the first two of those questions.
What the Harvard researchers measured
The study, led by researchers at Harvard Medical School and Brigham and Women's Hospital and published October 15, 2026 in JAMA, followed 950,000 physicians, physician assistants and nurse practitioners in the United States who used OpenEvidence over two and a half years. From January 2024 to June 2026 they submitted 270 million queries, Harvard Medical School reports; Fierce Healthcare's account of the paper gives 271 million.
The team used a large language model to classify the clinical purpose and topic of a random sample of 38.8 million queries. The National Institutes of Health supported the work. OpenEvidence did not fund it but supplied the de-identified data, and three of the co-authors are employees of the company.
What clinicians ask OpenEvidence day to day
Most queries, 51.9%, concerned clinical decision making, particularly treatment and management. Diagnosis accounted for 13.9%. Questions often fell outside the clinician's specialty: surgeons, for example, often asked about subjects other than surgery.
"Two key takeaways are that an enormous number of physicians are using AI for clinical care and that they are often using it in areas where they don’t have specific expertise." — Anupam Jena, corresponding author, Harvard Medical School
Use varied by specialty. Oncology and hematology specialists made 3.8% of queries while forming 1.9% of the workforce. Internal medicine doctors made 27.3% of physician queries and form 14.5% of the physician workforce. Radiology, orthopedics and ophthalmology used less than their share.
How fast use grew and how often clinicians returned
Monthly active users rose from 355 in January 2024 to 703,265 in June 2026. Queries per active user grew from 18 a month to 39.5, and month-to-month retention reached roughly 90% in 2026. By June, 92.5% of active users were returning users rather than brief adopters.
More than half of the users in the sample, 556,200, were physicians, and they averaged 37 queries per month of use. Early-career physicians and those in oncology, hematology, internal medicine and family medicine sent a higher-than-average share. Jena contrasted that speed with electronic health records, whose adoption was slow and, in his words, "only really occurred after government regulation."
What OpenEvidence says about its reach
The company's own figures go further, and they are its claims. As of October 2026, founder and CEO Daniel Nadler told Fierce Healthcare, 1.2 million medical license-verified US clinicians, including physicians and nurses, were using OpenEvidence, and they consulted it 45 million times in October alone. The company says roughly two-thirds of US physicians use the platform. Nadler also said it handles about one query from a medical specialist for every specialist visit in the country, against 300 million such visits a year.
What it changes for hospitals and health systems
The adoption came from individual clinicians, and the company is now moving into health system records. On October 29, 2026, Fierce Healthcare reported that NYU Langone Health had embedded OpenEvidence in its Epic electronic health record workflows. Clinicians there can ask clinical questions in natural language and get answers grounded in peer-reviewed literature and read against the patient's own record without leaving Epic, according to company executives.
The scale is large: NYU Langone has more than 12 million outpatient visits a year, 421,000 emergency department visits, more than 6,000 physicians and faculty, and 54,000 employees. OpenEvidence already has an enterprise rollout across NewYork-Presbyterian's 10 hospitals and nearly 350 care sites, a system-wide deployment at Mount Sinai and a precision oncology partnership with Memorial Sloan Kettering Cancer Center. For a health system, the task this handles is the evidence lookup at the point of care. Whether it improves patient care has not yet been measured; the researchers say that is the next step.
Questions and answers
How many doctors use OpenEvidence?
The JAMA study counted 703,265 monthly active users in June 2026, and 556,200 physicians in its sample. The company told Fierce Healthcare that 1.2 million license-verified US clinicians were using it as of October 2026.
What do doctors use AI for most?
In this study, clinical decision making, particularly treatment and management, made up 51.9% of queries, and diagnosis made up 13.9%. Many questions were outside the asker's specialty.
Who funded the study of OpenEvidence use?
The National Institutes of Health supported it. OpenEvidence provided de-identified data but no funding, and three co-authors work for the company.
MOZGII view
How we would build it
An evidence tool that clinicians adopt on their own is the easy half. The hard half is what a health system does once that tool sits inside the chart. If we were building it, we would keep two things apart: the medical literature, which is public and changes slowly, and the patient record, which is private and changes by the hour. Every answer would show which papers it relied on and which parts of the chart it read, so a doctor could check both in seconds.
The study shows that many questions fall outside the asker's specialty. That is where the risk sits, because the doctor is less able to catch a wrong answer. We would route those questions to a stricter mode: shorter answers, more citations and a clear note when the evidence is thin. We would also log each question by specialty and topic, the way the researchers classified theirs, so the hospital knows what its staff actually ask.
The one metric we would watch is not usage, which the study already shows is high. It is how often an answer changes a treatment decision, and whether that change holds up when a specialist reviews it. The researchers name that as their next question. Next, we would build a review queue where specialists sample answers in their own field each week and mark the ones they would not sign.
Sources
Where the facts in this story come from.