Clinicians using ambient AI scribes spent less recorded time writing notes in one US study, without a clear reduction in after-hours electronic-record work. The randomized crossover trial also found improvements in clinicians’ reported experience. Its product comparison and its comparison with the preceding baseline answer different questions.
The finding helps explain why a promise of “time back” needs a more precise question: time spent on which task, measured how, and compared with what?
Two products, tried by the same clinicians
The JAMIA study enrolled 160 outpatient clinicians who volunteered for a pilot. Participants were randomized to use two anonymized products in different orders, in periods lasting about a month during late 2024. Both tools were integrated with the electronic health record, or EHR.
There were 136 clinicians with at least one follow-up survey. Objective EHR measures had their own missing-data counts, so that survey denominator should not be applied to every result.
The crossover design is useful for comparing products: the same clinician can contribute experience of both. It was not a randomized comparison in which one group continued without a scribe. Changes from the pre-pilot baseline therefore need a different interpretation from the randomized product comparison.
What happened to recorded time
Table 3 separates the outcomes rather than combining them into a single efficiency score.
| Measure | Reported finding | What the comparison answers |
|---|---|---|
| Daily time in notes | Product A: −5.88 minutes; product B: −9.07 minutes from baseline | Changes during the pilot, compared with the preceding baseline |
| Product difference in daily note time | B minus A: −3.19 minutes; 95% confidence interval −4.87 to −1.50 | The randomized comparison between these two products |
| EHR time outside scheduled hours | Neither product had a statistically conclusive reduction from baseline | Whether the pilot demonstrated less after-hours EHR activity |
The 3.19-minute figure is particularly easy to misread. It is the estimated difference between products, not a universal saving from adopting AI and not a saving per consultation. The unit is minutes in notes per day.
Likewise, the absence of a conclusive after-hours reduction is not proof that no clinician saved time. It means that this study did not establish that improvement at the measured group level.
Experience improved without the same change in every clock
Clinicians reported greater workflow satisfaction with both products, with a larger improvement for product B. Reported personal and work-related burnout also fell relative to baseline, without a clear difference between products.
Those findings concern reported experience, not simply the duration of EHR use. They can coexist with limited changes in after-hours activity. The study’s authors discuss possibilities such as changes in the effort of composing notes and how saved time was used; those explanations were not separately established by the trial.
A purchasing decision would therefore lose information if it reduced the results to either “the tool saves time” or “the tool does not work”. The study measured several outcomes that matter for different reasons.
The limits of a short pilot
Participants knew which product they were using, had volunteered early for limited places, and learned two systems over a short period. There was no scribe-free washout between products. The timing included the Thanksgiving period, software features changed, and the analysis did not adjust for multiple testing.
The study also collected free-text safety concerns. Those replies do not substitute for an independent audit of note accuracy or patient outcomes.
For teams reading scribe evidence, the useful comparison starts with separate measures: note time, total EHR work, work outside scheduled hours, clinician experience and documentation quality. This trial provides evidence about some of those measures in one setting. It does not turn a single time estimate into a complete account of the working day, or a ranking of today’s vendors.