What 2026 research says about AI scribes and your evenings

This spring, two studies took a close look at AI scribes. One measured the time they save. The other graded the notes they write. The short version: scribes help, but they don't take the final read off your plate.

The daytime savings are real

In April, JAMA published a study of 8,581 clinicians at five academic health systems. Of those, 1,809 adopted an AI scribe. Adoption was linked to 16 fewer minutes of documentation for every 8 scheduled hours of patient care. Total EHR time dropped 13.4 minutes on the same basis. Adopters also saw about half an extra visit per week. The authors called these changes modest.

Modest still counts. If you have 32 scheduled patient hours a week, that works out to roughly an hour less documentation. Primary care clinicians were among the groups with the largest changes. So were clinicians who used the scribe in half or more of their visits.

Evenings mostly stayed the same

The same study found that EHR time outside work hours "did not change significantly." For many clinicians, that is the number that matters most. The time saved did not show up as less EHR time after hours.

The study measured time, not reasons. But a scribe only drafts the note. Inbox messages, refills, results, and the final read before signing still land on you.

The draft still needs a careful read

Also in April, the Annals of Internal Medicine published a Veterans Health Administration comparison of notes. Five primary care cases with standardized patients were audio recorded. Eleven AI scribe tools and 18 human note takers wrote notes from the same audio. Thirty blinded raters scored the notes on a 50-point quality scale.

Human notes scored higher in all five cases. AI notes scored lower in all ten quality areas. The biggest gaps were in being thorough, organized, and useful. The widest case gap was acute low back pain: 43.8 for human notes and 20.3 for AI notes.

To be fair to the scribes, the cases were simulated. The human note takers also weren't writing under real clinic pressure. Even so, the authors concluded that "independent, vendor-neutral evaluations of note quality are essential before large-scale clinical deployment."

Payers are watching the codes, too

Scribes may affect billing as well as notes. Trilliant Health, a health data firm, studied 2018 to 2024 claims from six large health systems that had announced ambient AI scribes. At those systems, the share of established-patient visits billed as 99214 or 99215 rose 7 to 12 percentage points.

Trilliant's own read was measured. It said the rise warrants a look but likely reflects more complete documentation, not fraud. It also said "the underlying causes of the changes are not clearly understood." Insurers sound more worried. In March, the Blue Cross Blue Shield Association said AI coding tools at hospitals may be creating gaps between diagnoses and treatment.

For your practice, the point is simple. A fuller note can honestly support a higher level. But if a draft says more than happened in the room, any level built on it is wrong. The level and the codes should match what the note supports, up or down.

A short check before you sign

Keep the scribe if it helps you, and spend a few of the saved minutes on a real read. Before you sign, check four things:

1. Does the history say what the patient actually told you, with the right doses?
2. Does the assessment and plan show what you decided today for each problem?
3. Do the diagnosis codes match the assessment, including side and stage where they apply?
4. Does the visit level match what the note supports, whether that is higher or lower?

A biller can run the same check before the claim goes out. It catches mistakes in both directions.

Cortex Lens is built for the last two questions. It is a read-only Chrome extension that works inside Practice Fusion. It flags codes that can't be billed together, a missing left or right, two codes that should be one, and a missing or conflicting CKD stage. Your practice can also turn on a visit level check. Flags go both ways, and a person makes every change. You can see how it scores on our public benchmarks.