What changed in our 12-week pilot with 40 clinicians

From May to August, 40 clinicians in family medicine and pediatrics used Charted for every scheduled visit. We measured the time from the end of each visit to a signed note, using timestamps from their EHRs, and compared it with the four weeks before the pilot.

How we measured

We tracked one interval: from the end of the visit to the clinician's signature. Both timestamps already live in the EHR, so nobody had to keep a log or estimate their own time. The interval also takes in everything that happens to a note after the patient leaves, including review and edits. A note signed at home at 10 p.m. counts in full.

We report medians rather than averages. A handful of notes left until the next morning can pull an average far away from a typical visit, and the median is closer to what a clinician sees on an ordinary day. The baseline is the four weeks before the pilot, timed the same way.

What changed

The median time to a signed note dropped from 14 minutes to 4, about 70% less. Same-day sign-off rose from 61% to 97%. Before the pilot, about four notes in ten were finished on a later day. During the pilot, it was about three in a hundred.

The two results are connected. The draft is ready within a minute of the visit ending, so the work left is review, and review is short enough to finish before the next patient. As I wrote in What to check first when you review an AI-drafted note, the clinicians who signed fastest made a habit of that: they signed each note before opening the next chart.

Where the edits went

Clinicians edited a median of three lines per note, so half of all notes needed three changed lines or fewer. Most edits were in the plan, where clinicians added details they hadn't said aloud. A draft can only contain what was said in the room. If you decide on a follow-up interval but never say it, Charted has nothing to draft it from.

The simplest fix we found was saying the plan out loud before the patient left. Silent exam findings won't be in the draft either, and Charted flags those sections for you. For codes, clinicians changed about one ICD-10 suggestion in eight. Grace Holloway's Reviewing AI-suggested ICD-10 codes without slowing down covers how to check those quickly.

What stayed the same

Visits took about as long as before. Recording doesn't make the conversation shorter, and the time clinicians got back came after the patient had left.

Every patient was asked before recording started. About 4% declined, most often at a first visit or when the visit turned to something sensitive. Two clinicians also chose not to record sensitive visits, and we think that was the right call. A clinician can pause with one tap, and the note records whether consent was given and whether the recording was paused. When a visit isn't recorded, the clinician writes the note the usual way.

If you run your own pilot

• Pull a baseline from your EHR before you start. We used four weeks.

• Time each note from visit end to signature, and track the share signed the same day. Both come from timestamps your EHR already keeps.

• Choose a group with full schedules. Health systems often start with 10 to 20 clinicians before a wider rollout.

• Watch where the edits land. If most are in the plan, coach the habit of saying the plan aloud before you change any templates.

• Count declines and paused recordings alongside the time.

We help clinics choose the group and measure the results.

Questions we get

Can I read the full report?

Yes. The full method and results are in the pilot report, which we share on request.

Why do the specialty pages show different numbers?

Each specialty page reports a single clinic's own 12-week pilot. This post pools 40 clinicians in family medicine and pediatrics, so its totals won't match any one clinic's page, and clinics in other specialties aren't part of them.