JAMA study counts 952,008 US clinicians and 271 million queries on OpenEvidence in 2.5 years

A new cross-sectional study in JAMA puts hard numbers on clinical AI adoption: between 1 January 2024 and 30 June 2026, 952,008 US clinicians with verified National Provider Identifier accounts sent 271 million queries to OpenEvidence, an AI-enabled clinical information platform¹.

What the study found

Growth was steep. Monthly active users rose from 355 in January 2024 to 703,265 in June 2026, and the average number of monthly queries per active user more than doubled, from 18.0 to 39.5¹. Of the clinicians who used it, 556,200 were physicians¹.

Physicians were not the heaviest users per head. Nurse practitioners and physician assistants each averaged 44.5 queries per active user-month, against 37.1 for physicians¹. Measured against their share of the national workforce, physicians within ten years of graduation submitted 2.54 times their expected share of queries, and residents and fellows 1.84 times¹.

Specialty mattered a great deal. Use ratios were high in oncology and haematology (2.00), internal medicine (1.88) and family medicine (1.80), and very low in procedural and diagnostic fields: ophthalmology at 0.23, and radiology and orthopaedics at 0.24 each¹.

The researchers had a large language model classify a stratified random sample of 38.8 million queries. Just over half (51.9%) concerned clinical decisions, with treatment and management alone accounting for 33.1%; diagnosis made up 13.9%¹. A large share of questions fell outside the asking physician's own field, from 35% among oncologists and haematologists to 88% among emergency physicians¹.

What it means for the market

For buyers and investors, this is rare independent usage data on a clinical AI product in place of vendor-reported figures. It shows a tool reaching a large part of the US clinical workforce, with demand concentrated among non-physician clinicians, younger doctors and cognitive specialties¹.

The usage pattern also says something about where the product fits. Most queries were about what to do for a patient rather than what the patient has, and many crossed specialty lines¹. On this evidence the platform is being used chiefly as point-of-care decision support rather than as a diagnostic aid, and it has made little headway in radiology, ophthalmology and orthopaedics¹.

What the abstract does not show is equally relevant to the market: it reports no data on revenue, on outcomes for patients, or on the accuracy of the answers clinicians received¹. Whether heavy use translates into better or safer decisions, or into contracts with health systems, is not addressed.

Caveats

The study is cross-sectional and covers one platform and US clinicians only¹. Query purposes were labelled by a language model working on a sample rather than read by people, and the analysis measures how often clinicians asked questions, not what they did with the answers¹.