JAMA Netw Open
AI hospital summaries used safely, cut physician burnout
Clinical Takeaway: AI-drafted hospital summaries may ease documentation strain and burnout when doctors review them carefully. Missing details remain the main concern.
Writing discharge summaries is one of the more time-consuming tasks in hospital medicine and a major driver of burnout. Earlier studies have evaluated the quality of AI-written summaries, but this study tested whether an AI tool could safely fit into the daily workflow of inpatient care.
Across 384 discharges, physicians used AI-generated text in their final note 57% of the time. Of 100 summaries with detailed safety feedback, 88% were rated as posing no potential for harm and just one (1%) was rated as likely to cause moderate harm; none were judged likely to cause severe harm or death.
Average burnout scores dropped from 1.75 before the pilot to 1.20 after, on a 0-to-4 scale where higher scores mean more burnout. The group started above 1.33, the standard cutoff for clinically significant burnout, and ended below it.
The errors that did appear followed a pattern. The most common problem was missing information (25%), followed by factual mistakes (20%). Made-up content was rare (2%), and citation errors were never reported.
Many of the missing items reflected events that happened after the overnight draft was generated, lab and pathology data stored outside clinical notes, or details the model didn't recognize as important.
Time savings were small, with five of seven physicians shaving up to 2.9 minutes off median documentation time. That was not statistically significant, but two-thirds of physicians said they felt the tool saved them time and pointed to mental relief rather than faster typing as the real gain.
The tool, called MedAgentBrief, ran on Gemini 2.5 Pro and built each summary in three steps: drafting, refining, and checking for hallucinations. Each statement was linked back to its source note. Drafts were emailed overnight to attending hospitalists, who could use, edit, or ignore them. The 10-week pilot ran on a general medicine inpatient unit at Stanford Health Care in Redwood City, CA, which is staffed by 11 attending hospitalists.
For hospitals considering AI documentation tools, the real payoff may be in reduced physician mental load and burnout, not minutes saved. The authors warn that if those gains get translated into pressure for higher patient volumes, the benefit could disappear.
"These findings suggest that agentic LLM workflows with reasoning models may be integrated into active clinical care with minimal safety risk and may be associated with reduced physician burnout, even when objective time savings are modest," the authors concluded.
Source: Grolleau F. JAMA Netw Open. 2026 May 8. Physician-Reported Safety Outcomes of AI-Generated Hospital Course Summaries