AI Discrepancy: Patients Ask AI for Diagnosis. Doctors Use It for Treatment.

JAMA analyzed 271 million real clinical queries from 952,008 US clinicians on OpenEvidence (Jan 2024 - June 2026). 

51.9% of queries were for clinical decision-making, particularly treatment and management (33.1%), while diagnosis was only 13.9%.

Who's actually using it?

Career stage is the biggest driver:

<10 years from med school: 24.3% of workforce, 61.8% of queries — ratio 2.54

Residents/fellows: 14.8% of workforce, 27.2% of queries — ratio 1.84

20+ years out: 53% of workforce, only 16.5% of queries — ratio 0.31

An "inversion of expertise" — trainees are more fluent than attendings. 90% monthly retention by 2026. This is habit, not experimentation.

Specialty:

Highest: Oncology/Hematology (2.00), Internal Medicine (1.88), Family Medicine (1.80)

Lowest: Ophthalmology (0.23), Radiology (0.24), Orthopedics (0.24), Anesthesiology (0.37)

Fast-moving, cognitive fields lead. Procedural fields lag.

Intensity:

NPs and PAs: 44.5 queries per active user-month vs Physicians: 37.1 — and they were >40% of users.

Questions outside the specialty:

Surgeons: 70% of queries outside surgery

Anesthesiologists: 73% outside anesthesiology

Even oncologists: 35% outside oncology

References:

https://jamanetwork.com/journals/jama/fullarticle/2855491