Peer-reviewed research
Quality of Clinical Notes Created by Ambient Listening Generative AI: Pragmatic Prospective Pilot Study
Taylor SL et al., JMIR Medical Informatics, 2026 · DOI 10.2196/86474
Read the originalRead on October 16, 2026
- What it is
- A quality improvement pilot at the University of California, Davis, in July and August 2024. Thirty-one physicians used an ambient listening AI scribe to create 7,545 clinic notes, and they reviewed 356 of those notes for four kinds of error: accidental inclusions, accidental omissions, hallucinations and bias.
- How much weight it carries
- Peer-reviewed, and run in real clinics rather than in simulation. Physicians reviewed only 4.7 percent of the notes, the study has no control group, and it measures how often errors occurred, not time saved or how the notes compare with notes written by hand.
Facts taken from this source
- In the UC Davis pilot, hallucinations, meaning content the recording did not support, appeared in 11.5 percent of the 356 reviewed notes, the second most common error after omissions.
“accidental omissions were the most frequent error (n=64, 18%), followed by hallucinations (n=41, 11.5%), and accidental inclusions (n=33, 9.3%)” (Results)
- In the UC Davis pilot, accidental inclusions, meaning content from the visit that does not belong in the note, appeared in 9.3 percent of the 356 reviewed notes.
“accidental omissions were the most frequent error (n=64, 18%), followed by hallucinations (n=41, 11.5%), and accidental inclusions (n=33, 9.3%)” (Results)
- In a 2024 pilot at UC Davis, 94.7 percent of the 356 AI-generated clinical notes that physicians reviewed were free from significant errors.
“AI-generated clinical notes were generally of high quality, with 94.7% (337/356) free from significant errors.” (Results)
- In the UC Davis pilot, accidental omissions were the most frequent error, in 18 percent of the 356 reviewed notes, ahead of hallucinations at 11.5 percent and accidental inclusions at 9.3 percent.
“accidental omissions were the most frequent error (n=64, 18%), followed by hallucinations (n=41, 11.5%), and accidental inclusions (n=33, 9.3%)” (Results)
- In the UC Davis pilot, 19 of the 356 reviewed notes (5.3 percent) contained an error that physicians rated as posing serious or imminent risk if it was not corrected.
“only 19 (5.3%) notes containing errors rated as posing serious or imminent risk (severity 4-5)” (Results)
Pages that rest on this source
- Meeting Transcription: How It Works and What to UseGuide
- AI Cleanup for Dictation: Fillers, Punctuation, RewritesGuide
- AI Medical Scribes and Ambient Clinical DocumentationGuide
- Is Dictation HIPAA Compliant? What Makes It SafeGuide
- AI Voice Notes: From Rambling to Summaries and To-DosGuide
- Are AI summaries of voice recordings accurate?Question
- How accurate are AI medical scribes?Question
- How do you turn a meeting transcript into minutes?Question
- What is the difference between a meeting transcript and meeting minutes?Question
- Should doctors use an AI scribe or dictation?Question
- What is an ambient AI scribe?Question
- Can ChatGPT summarize a voice memo?Question
- Does AI dictation change what you said?Question
- How do you turn voice notes into a to-do list?Question
- What typing speed do you need to be a medical scribe?Question