Stanford Health Care rolled out an ambient AI scribe in its emergency department (ED) in stages between February 2025 and March 2026, and a retrospective cohort study of 10,344 encounters managed by 100 attending physicians found that using it was associated with a 72.6-second reduction in on-shift documentation time per encounter (95% CI 63.8 to 81.4; P<.001). The adjusted estimate corresponds to a 28% decrease from the model-predicted baseline, from 260 to 187 seconds, or roughly 24 minutes saved per 8-hour shift if used across 20 encounters. Adoption reached 18.2% of encounters (1,881 of 10,344).
The authors, including Carl Preiksaitis of Stanford's Department of Emergency Medicine, write in JMIR AI that "To our knowledge, this is the first inferential evaluation of ambient AI scribes on documentation time in the ED using EHR audit log data."
Clinician burnout has reached crisis levels in emergency medicine, with documentation burden identified as a central contributing factor, and the ED's high patient acuity, rapid turnover, constant task switching and 24/7 operations amplify it. Earlier evidence on ambient scribes came mostly from outpatient and primary care settings. A prior descriptive analysis from the same institution, covering the first 8 months of the ED rollout, found low but highly skewed uptake (11.2% of encounters and 38% of attending physicians, clustered in low-acuity zones) and an unadjusted median documentation time about 65 seconds shorter for ambient encounters.
The rollout ran from February 1, 2025, to March 1, 2026, with voluntary adoption; attending physicians decided encounter by encounter whether to use the scribe. Stanford Health Care collaborated with Epic Systems Corporation and Nuance Communications to integrate Microsoft DAX Copilot ambient AI scribe technology into clinical workflows. Clinicians record patient-clinician conversations from the Epic Haiku mobile app, and draft note content is parsed into four auto-populated Epic SmartSections: history of present illness, physical examination, results, and assessment and plan. An attestation SmartSection documents patient notification and clinician review, and training resources, including knowledge base articles, presentations and individual or group sessions, were available to physicians.
The analysis covered four ED care settings and was restricted to encounters managed by a single attending physician without human scribes, excluding the resuscitation and pediatric zones. On-shift documentation time came from Epic's User Action Log audit data with a 5-second inactivity time-out, and the team used mixed-effects linear models with physician random intercepts, plus negative control outcomes, a within-clinician placebo permutation test (mean -0.8 seconds), sensitivity analyses and doubly robust analyses.
The effect was similar for high-use physicians (use rates of at least 18.2%; -71.6 seconds) and low or moderate users (-64.2 seconds), with no statistically sign