Ambient AI scribes: one year and 2.5 million uses, in medicine rather than dentistry
The Permanente Medical Group measured a year of generative-AI scribe use across 7,260 physicians. It is the largest public rollout we could find. It is also medical, so before any dental owner treats it as a benchmark, the gap between a primary-care visit and a hygiene or restorative appointment has to be spelled out.
Written Oct 31, 2026 (ET) Three secondary pages and ADA HPI fetched Oct 31, 2026 · NEJM Catalyst primary returned 403
For these physicians, mostly yes: less after-hours charting, and heavy users gained the most. The evidence we read did not test transfer to the dental chair.
- 1 Vendor case study
- 2 Operator-reported
- 3 Organization evaluation
- 4 Randomized trial
What operators / vendors claim
Reported by the named group or vendor. Treated as claims until independently verified.
- Time saved. As reported by the AMA and Permanente, the NEJM Catalyst analysis estimates 15,791 hours of documentation time saved over one year, described as "1,794 working days", compared with nonusers. Claim · organization-reported
- Physician sentiment. Permanente's release says 84% of physicians reported a positive effect on patient interactions and 82% on overall work satisfaction. Claim · survey
- Patient sentiment. 47% of patients said their doctor spent less time looking at the computer, and the AMA reports that "no patients reported a negative" impact on visit quality. Claim · survey
- Heavy users benefit most. The top third of users accounted for 89% of activations, and high users saved "two and a half times more time per note" than less frequent users (Permanente). The AMA phrases this as "more than double". Claim · organization-reported
What we independently checked
What Signal Desk could confirm from public sources this pass — not a clinical study.
- Scale and window. Kaiser Permanente's Division of Research states that between October 2023 and December 2024, 7,260 TPMG physicians used the tool in 2,576,627 patient encounters, and 3,447 physicians used it in at least 100 encounters. Checked · KP Division of Research
- What the tool does and does not do. Every page we fetched says the scribe drafts notes from the conversation and does not make care decisions. Physicians review and edit the draft, and patients are told it is in use. Checked · sources agree
- Accuracy sample. The Division of Research page says the earlier pilot paper scored 35 AI-generated transcripts at an average quality score of 48 on a 50-point scale, with "few incidents of hallucinations". Checked · KP Division of Research
- The vendor changed mid-study. The organization switched scribe vendors between the 2023 pilot and the 2024 rollout, and adoption kept rising. Checked · KP Division of Research; AMA
- Dental demand signal. The ADA Health Policy Institute (July 2026) reports that 34.8% of responding dentists not yet using AI plan to use it for charting and note taking. Checked · ADA HPI
What did not go well / limits
Gaps, caveats, and anything we could not verify. Numbers without a fetched source stay here.
- Primary paper not opened. NEJM Catalyst returned a 403 to our fetch, so we have not read the methods, statistical models, or the authors' own limitations section. Unchecked · primary paper
- Secondary sources disagree slightly. The Division of Research page reports 88% of 102 surveyed physicians saying the scribe had a positive impact on visit interactions. Permanente and the AMA report 84%. For patients, one says "more than a third" and the others say 39%. These may be different questions or samples, and we cannot tell which without the paper. Limit · inconsistent restatements
- Users chose themselves. Physicians opted in, and heavy users already spent more time on notes before the rollout. Comparisons with nonusers are not randomized. Limit · selection
- Small surveys. 102 physician responses and 118 patient responses, out of thousands of users. Limit · sample size
- Medicine is not dentistry. Dental notes involve tooth-level charting, perio numbers dictated over instruments, and hygienist-led visits. None of that was in the studied workflows, and the vendors and costs were not named in what we read. Limit · transferability
- Possible affiliation. The AMA article notes that TPMG is part of the AMA Health System Member Program. That does not invalidate the coverage, but it is not an independent review. Limit · secondary-source independence
Owner takeaway: before signing up for a dental scribe, ask for dental-specific accuracy samples (perio charting and tooth numbering), the exact way patients are told it is in use, and a 30-day check of after-hours charting time on your own team.
Sources
- Ambient Artificial Intelligence Scribes: Learnings after 1 Year and over 2.5 Million Uses
- AI-assisted notetaking gains steady support from Kaiser Permanente physicians
- Analysis: AI scribes save physicians time, improve patient interactions and work satisfaction
- AI scribes save 15,000 hours—and restore the human side of medicine
- Dentists Use AI to Make Appointments More Efficient, but Draw the Line at Clinical Decision Making
← All case studies · Related directory category · How we label evidence · Editorial case for practice owners evaluating AI tools. Not a vendor advertisement.