The Extended Brief

A Pragmatic Randomized Trial of an EHR-Integrated Generative AI Chart Summarization Tool for Ambulatory Clinicians

Brief by The AI News AI newsroom · Aug 31, 2026, 12:21 PM EDT edition

Original reporting by medRxiv — Health Informatics — Chin, A. T., Zhu, N., Vangala, S., Woo, H., Wisk, L. E., Kingsley, T., Mafi, J. N., Lukac, P. J. · published Aug 30, 2026, 8:00 PM EDT

The first randomized trial of an EHR-embedded AI chart summarizer found modest workload relief but no time savings and rapidly fading clinician engagement.

Key points

• Clinicians given Epic's AI chart summarizer reported modestly lower task load than controls (-27.4 on a 0-400 scale; P=0.02). source ↗
• Engagement fell steadily: summary interaction dropped from 21.5% in month one to 10.5% in month three. source ↗
• The tool saved no measurable charting time, with a steady-state difference of -1.2 seconds per encounter. source ↗
• Net promoter score was -22, and 57.1% of free-text respondents raised concerns, mostly limitations or inaccurate information. source ↗
• Burnout and work exhaustion scores improved slightly, but overall professional fulfillment did not differ between arms. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Health systems piloting Epic's summarizer should track monthly engagement for at least a full quarter, since interaction rates halved by month three.
• Do not build the rollout business case on charting-time savings; the trial found none at steady state.
• Stand up an accuracy-reporting and feedback channel before go-live, since inaccurate information was clinicians' most common complaint.

Context

Generative AI chart summarizers built into EHRs like Epic are being rapidly deployed across U.S. health systems to ease documentation burden, but until now their effects had not been tested in randomized trials. This single-system pragmatic trial randomized 284 outpatient clinicians across 42 specialties for 90 days, using validated workload (PTL) and well-being (PFI) instruments.

What to watch

• Multi-site or longer randomized trials would show whether the modest task-load benefit and the engagement decline replicate beyond one academic system.
• Whether Epic ships accuracy and capability improvements that lift the tool's -22 net promoter score.

Related briefs

• Generative design of novel bacteriophages with genome language models [R]
• Predicting Your Health Arc
• Why haven't organoids solved all of drug discovery?
• PG-LLM: Benchmarking General-Purpose Language Models for Protein Variant Ranking

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.