Ambient AI scribes in clinical practice, and what the trials show

Randomised trials in NEJM AI and multisite studies in JAMA and JAMA Network Open show that ambient AI scribes reduce the time doctors spend on notes and lower measures of burnout, with modest effects that vary by product, specialty and how often the scribe is used. They also report occasional clinically significant inaccuracies, which is why every note still needs a doctor's review before it is signed.

The evidence has grown quickly. Two randomised trials were published in NEJM AI in November 2025, and a multisite study of 8,581 clinicians at five US academic health systems appeared in JAMA in April 2026.¹²³

Time spent on notes

In a three-group randomised trial at UCLA Health, 238 outpatient physicians across 14 specialties were assigned to one of two scribes or to usual care. Physicians using one scribe cut their time in notes by 9.5% compared with the control group, while the other scribe produced no significant change.¹

A 24-week stepped-wedge trial at the University of Wisconsin randomised 66 practitioners. Of 71,487 notes, 38% were written with ambient AI, and time spent on notes fell by 0.36 hours a day, about 22 minutes.²

The JAMA study compared 1,809 adopters with other clinicians at the same institutions. Adoption was associated with 16.0 fewer minutes of documentation time and 13.4 fewer minutes of total record time per eight scheduled patient hours, and 0.49 additional visits a week. The gains were largest in primary care and among clinicians who used the scribe in half or more of their visits.³

In an earlier pre-post study of 46 clinicians at the University of Pennsylvania, time in notes fell from 10.3 to 8.2 minutes per appointment and after-hours work fell from 50.6 to 35.4 minutes per workday.⁴

Uptake was partial even inside the trials. In the UCLA study, the two scribes were used in 33.5% and 29.5% of eligible visits.¹ In the Wisconsin trial, the reduction in work outside working hours was no longer significant once the most extreme 3% of daily observations were removed.² The JAMA study found larger changes for advanced practice clinicians and female clinicians as well as for primary care.³

Burnout and attention

In the Wisconsin trial, work exhaustion and interpersonal disengagement fell by 0.44 points on a five-point scale, while the rise in professional fulfilment was not significant.² In a randomised crossover trial of 160 clinicians at Duke, both scribes tested reduced personal and work burnout scores.⁶

A quality improvement study across six US health systems found that the share of 263 clinicians reporting burnout fell from 51.9% to 38.8% after 30 days with a scribe, and clinicians reported more ability to give patients their undivided attention.⁵ The study had no control group, and two of its authors worked for the scribe's vendor.⁵

What changes in the consultation

The Pennsylvania study found same-day appointment closure rose from 66.2% to 72.4%, and clinicians reported feeling more engaged with patients.⁴ In the Duke trial, satisfaction with workflow rose by 2.51 points on a seven-point scale with one product and 1.91 points with the other, and the first product saved 3.19 more minutes in notes a day.⁶

Accuracy and safety

In the UCLA trial, clinicians reported clinically significant inaccuracies occasionally with both scribes, scoring about 2.7 to 2.8 on a five-point scale, and one mild adverse event was reported. The authors concluded that the inaccuracies require ongoing vigilance.¹

In the Wisconsin trial, documentation quality scored between 3.97 and 4.99 across domains on a five-point instrument, diagnostic billing codes improved, and no drift in software performance was detected.²

Singapore's AIHGle 2.0 guidelines treat a scribe that summarises the consultation for the doctor's review as a human-in-the-loop clinical operations tool at negligible to minor risk, and state that human oversight is mandatory.⁷

What the evidence does not show yet

All six studies cited here come from US health systems, the trials ran for weeks to months, and several outcomes are self-reported. The JAMA study found no significant change in record time outside working hours, and the Duke trial found after-hours time largely unaffected.³⁶

Results also differed between products in both head-to-head trials.¹⁶ A clinic should therefore test the tool it plans to buy in its own consultations and its own languages before committing. None of the studies cited here was run in Singapore, so performance in consultations that move between English, Mandarin, Malay or Tamil has to be tested locally.

Questions to ask before buying

Which published studies used this product, and what did they measure?
Does it recommend a diagnosis or treatment, which would make it a medical device in Singapore?
How does a doctor flag an inaccurate note, and what happens next?
Where are recordings, transcripts and notes hosted, and for how long are they kept, in writing?
Does it work in the languages your patients use in the room?

wai hosts your clinic's records in Singapore, under Singapore's rules. We set this up with your clinic.

How to run a pilot in a Singapore clinic

Measure each doctor's time on notes and after-hours work for two weeks before the pilot.
Run the scribe for at least four weeks, with each participating doctor using it in most visits, since benefits were larger at higher use.³
Review a sample of notes each week against the consultation and log every inaccuracy.
Record the tool in the clinic's AI registry and agree the patient notice, as AIHGle 2.0 asks.⁷
Compare time and accuracy at the end, and decide with the doctors who used it.

What wai Clinicians does

wai Clinicians writes the note while the doctor talks and drafts the letter before the patient has left the room, and nothing is sent until the doctor has read it and signed it. The change for the doctor is a consultation spent looking at the patient, and the day's notes finished before they leave.

Questions

Do AI scribes save doctors time?

Yes, by modest amounts that vary by product. A JAMA study of 8,581 clinicians found adoption was associated with 16 fewer minutes of documentation per eight scheduled patient hours.³

Are AI medical scribes accurate?

Not always. In a randomised trial, clinicians reported clinically significant inaccuracies occasionally with both scribes tested, so every note needs a doctor's review.¹

Do AI scribes reduce physician burnout?

Randomised trials found reductions in work exhaustion and burnout scores.²⁶ Effects on after-hours work were smaller or not significant.

Is an AI scribe regulated in Singapore?

AIHGle 2.0 treats a scribe that summarises the consultation for the doctor's review as a negligible to minor risk tool with mandatory human oversight. A scribe that recommends a diagnosis or treatment is regulated as a medical device.⁷

Sources

Written and reviewed by the clinical team at wai. Last reviewed 7 September 2026.