News Research Precision Endocrinology Personalized Treatment Ethics, Regulation, and Responsible Use

Study examines physicians' use of clinical AI

November 09, 2026 By Julie Greenbaum 7 min read
Share Share via Email Share on Facebook Share on LinkedIn Share on Twitter

Physicians submitted more than twice as many queries about treatment and management as about diagnosis to an artificial intelligence–enabled clinical information platform, according to a cross-sectional study involving more than 950,000 US clinicians.

Early-career physicians and those specializing in oncology and hematology, internal medicine, and family medicine submitted disproportionately large shares of queries relative to their representation in the national physician workforce.

The study, published in JAMA, analyzed 271 million queries submitted to OpenEvidence, an artificial intelligence (AI)-enabled clinical information platform, between January 1, 2024, and June 30, 2026. Researchers found that monthly active users increased from 355 to more than 703,000 during the study period, while the average number of monthly queries per active user more than doubled.

Study design and methods

Researchers analyzed query logs from 952,008 US clinicians with verified National Provider Identifiers, including 556,200 physicians, 171,034 nurse practitioners, 93,241 physician assistants, 35,819 pharmacists, and 95,714 other licensed clinicians.

The platform uses large language models to synthesize responses to clinical questions, incorporating citations from peer-reviewed journals and professional society guidelines.

To characterize patterns of use, researchers examined query volume and intensity across clinician types, physician specialties, and career stages. They compared physicians' share of submitted queries with their representation in the national physician workforce.

A stratified random sample of 38.8 million queries, representing 14% of all queries, was classified by a large language model according to clinical purpose and medical topic. 

Researchers also evaluated classification agreement by comparing model-generated labels with assessments from 6 clinicians.

Clinical AI platform use increased over time

Physicians accounted for 58% of platform users and submitted 172.1 million queries, representing 64% of all queries during the study period.

Monthly active users increased from 355 in January 2024 to 703,265 in June 2026, while monthly query volume increased from approximately 6,000 to 27.8 million.

Average monthly queries per active user increased from 18.0 to 39.5 over the same period. By June 2026, 93% of active users had previously used the platform, and month-to-month retention had reached approximately 90%.

Query intensity differed across clinician types. Nurse practitioners and physician assistants each submitted an average of 44.5 queries per active user-month, compared with 37.1 among physicians, 29.3 among pharmacists, and 19.9 among other licensed clinicians.

Use varied by career stage and specialty

Physicians within 10 years of medical school graduation accounted for 24% of the national physician workforce but submitted 62% of physician queries. Residents and fellows, who represented 15% of the physician workforce, accounted for 27% of physician queries.

In contrast, physicians who graduated at least 20 years earlier represented 53% of the physician workforce but submitted 17% of physician queries.

Usage also varied substantially across specialties. Physicians practicing oncology and hematology accounted for 2% of the workforce and 4% of queries. Internal medicine physicians represented 15% of the workforce and submitted 27% of queries, while family medicine physicians accounted for 13% of the workforce and 23% of queries.

Procedural and diagnostic specialties demonstrated lower use relative to workforce representation. Ophthalmologists accounted for 2% of the physician workforce but submitted approximately 0.4% of queries. 

Among active physician users, average monthly query intensity ranged from 51.7 in family medicine and 49.7 in oncology and hematology to 19.9 in ophthalmology and 18.5 in orthopedics.

Treatment questions exceeded diagnostic queries

Among physician queries, clinical decision-making accounted for 52%, followed by clinical knowledge (37%), evidence (7%), administrative tasks (3%), and medical education (1%). Treatment and management represented the largest individual query category at 33%, followed by medication information (14%) and diagnosis (14%).

Query purposes differed across specialties. Medication information accounted for 29% of queries from psychiatrists, compared with 14% among physicians overall. Queries concerning specific studies and clinical trials were more frequent among oncologists and hematologists (4% vs 1%), while practice guideline questions were more common among gastroenterologists (4%) and cardiologists (4%) than among physicians overall (2%).

Differences also emerged by years since medical school graduation. Treatment and management queries accounted for a smaller proportion of queries among physicians who graduated at least 20 years earlier than among those who graduated fewer than 10 years earlier (30% vs 35%), whereas literature searches (5% vs 3%), risk and prognosis (4% vs 2%), and medication information (16% vs 15%) were more common.

Physicians frequently queried outside their specialties

Among emergency medicine physicians, 88% of queries addressed topics outside their specialty. The corresponding proportions were 70% among surgeons and 73% among anesthesiologists. Even among oncologists and hematologists, who had the highest proportion of within-specialty queries, 35% addressed other specialties.

Within-specialty queries were more frequently related to evidence than outside-specialty queries (9% vs 6%), including literature searches (5% vs 4%) and clinical practice guidelines (3% vs 2%).

Outside-specialty queries were more frequently directed toward clinical knowledge (37% vs 28%), particularly disease information (14% vs 9%) and medication information (11% vs 9%).

Study limitations

The researchers acknowledged several limitations, including reliance on data from a single AI-enabled platform, which may limit the generalizability of the findings. The analysis captured only queries submitted through the platform's main question interface, excluding other product features and institution-specific deployments. Query classification relied on a large language model rather than human annotation, potentially introducing misclassification. In a blinded validation, agreement between the model and clinician annotators was 74% for query purpose and 55% for medical topic, compared with 66% and 53%, respectively, between clinicians.

The researchers also could not determine whether queries occurred during patient encounters, clinical preparation, or nonclinical activities. The study could not establish whether AI use replaced traditional information sources or whether cross-specialty queries substituted for or preceded specialist consultations.

Additionally, because OpenEvidence holds content partnerships with several medical publishers, use patterns may not generalize to tools without comparable content access, and cross-specialty querying in particular may partly reflect the breadth of available content.

“For endocrinologists, these findings raise a practical question: how can AI-supported information retrieval help clinicians evaluate treatment options and apply evidence to individual patients?” said lead study author Vishal R. Patel, MD, of Harvard Medical School and Brigham and Women’s Hospital. “Our study describes use of one platform and cannot establish that using it improves clinical decisions or patient outcomes. Those benefits need to be evaluated directly, including in endocrinology, alongside the accuracy and appropriateness of the information clinicians receive.”

“Among physicians, treatment and management accounted for 33.1% of queries, compared with 13.9% for diagnosis,” Dr. Patel said. “This suggests that an important part of how clinicians use these tools is to seek information about managing patients, beyond identifying a diagnosis.”

Dr. Patel reported receiving grant support from the National Institutes of Health. Dr. Finlayson and Dr. Ray reported employment with OpenEvidence, which provided study data and contributed to the analysis. Dr. Zack reported serving as chief medical officer of OpenEvidence. Dr. Liu-Hass reported consulting fees from Accompany Health outside the submitted work. Dr. Jena reported financial relationships with several organizations and media outlets outside the submitted work and serving as an unpaid board member with the United Network for Organ Sharing. No other disclosures were reported.

AACE Endocrine AI is published by Conexiant under a license arrangement with the American Association of Clinical Endocrinology, Inc. (AACE®). The ideas and opinions expressed in AACE Endocrine AI do not necessarily reflect those of Conexiant or AACE. For more information, see Policies.

Related Content