The Effect of Ambient AI Documentation on Clinician Workload, Efficiency, and Patient Experience in a Multisite...

Evaluation of ambient AI on patient experience, documentation efficiency, clinician workload, and clinical throughput across a large emergency department (ED) network.

Retrospective, observational cohort study of ambient AI rollout across 14 EDs (May 2024-June 2025). Clinicians applied the tool on a voluntary basis and contributed both AI-assisted and conventional notes. Outcomes included patient-reported experience, active editing time, disposition-to-completion time, copied-forward content, and on-time note completion. Clinician workload was assessed with pre- and post-implementation surveys including the NASA Task Load Index (TLX).

Ambient AI was used in 8.6% of 315,242 ED clinical notes. AI use was associated with higher top-box ratings for clinician listening (82.0% vs. 76.6%; OR 1.39; p = 0.003) but not likelihood to recommend (77.6% vs. 75.0%; OR 1.15; p = 0.185). Within-clinician paired analysis showed no difference in active editing time between AI and conventional notes (median difference +0.17 min; 95% CI, -1.00 to +1.55; p = 0.349), and clinicians typed 722 fewer characters per AI-assisted note. Documentation finalized 3.4 hours earlier for admitted and 6.0 hours earlier for discharged patients; the disposition-to-completion difference was +5.06 hours (p = 0.337), reflecting attenuation of early-adopter's advantage. AI-assisted notes were half as likely to include copied content (9.4% vs. 18.3%; OR 0.46; p < 0.001). NASA-TLX workload decreased by 40.2 points (95% CI 30.4-50.1).

Ambient AI documentation was associated with improved patient-reported listening, earlier note completion, reduced copied-forward content, and lower clinician-perceived cognitive burden, without altering active editing time per note. These findings suggest a scalable tool for reducing clerical burden in high-throughput emergency care, while introducing new responsibilities for clinicians to review, edit, and sign machine-generated notes.