Ambient artificial intelligence scribe implementation in inpatient setting.

Abstract

Ambient artificial intelligence (AI) scribes have shown promise to aid ambulatory clinicians with the documentation burden, but little has been shown about the inpatient experience. Understand inpatient physician experiences before and after ambient AI scribes implementation. Quality improvement pilot evaluation using utilization data, pre-post surveys, and electronic health record (EHR) metrics at a large healthcare organization in northern and central California. Ambient AI was implemented in acute care facilities in March 2025, with EHR data collected 2 months before and after implementation. We examined ambient AI usage, inpatient Time in Notes per day, NASA-TLX cognitive load, the mini-Z burnout question, and overall experience. McNemar, paired t-test, and paired Wilcoxon signed-rank tests were used. Fifty-nine physicians were included: 36 (61%) male, 28 (48%) adult hospitalists, and 20 completed both pre- and post-surveys. Monthly utilization remained relatively stable (4.2%-3.5% notes by ambient AI). Ambient AI was used to generate 12% (647/5350) of history and physical notes, 2% (518/245,921) of progress notes, and 6% (33/511) of consult notes, with overall utilization rate of 3% (1198/46,161) across all inpatient note types. Mean time in notes (SD) per day was unchanged, from 41.9 (25.59) to 41.6 (27.3) min (p = .77). Physicians requested additional functionality to "pull forward" or add onto previously generated notes. Ambient AI was used most consistently for inpatient consult notes, although overall utilization was low. Future development tailored to inpatient workflows is needed to increase adoption as organizations begin to expand ambient AI in this new clinical setting.