Fourteen years into running a practice, I can identify the hour of the day documentation took from me: the hour after the last patient, every day. That is the hour ambient AI scribes promise to return. In my practice, and in the pilot literature, the promise is largely kept. The question worth a clinician’s attention is what the tool does with the details while it is keeping it.
Where the time actually goes
Evaluations of ambient documentation in outpatient settings report documentation time falling by about a third to a half among high-volume users. The less discussed finding is that the saved time follows the schedule, not the intention. Practices that kept visit lengths and templates unchanged banked the time as capacity. Practices that wanted longer conversations had to redesign templates on purpose. The tool returns minutes. It does not decide what they are for.
The failure modes are familiar
After several months of daily use, the errors I watch for are not dramatic inventions. They are omissions written in a confident voice. A pertinent negative mentioned in the room and missing from the note. A laterality flip in a paired structure. A plan item recorded as decided when it was offered and deferred. Each is the kind of error a tired clinician makes at 7 p.m., which is precisely the point: the scribe does not remove the review step, it changes what the review step has to catch.
A practical rule from our practice: read the assessment and plan line by line, skim the history, and never sign a note about a patient whose conversation you cannot still hear in your head.
The parts that are not clinical, and are
Consent language, audio retention, and what the vendor may do with de-identified data are procurement questions with clinical consequences. Two products with identical note quality can differ completely on whether recordings are retained, where they are processed, and what trains future models. That comparison belongs in the purchase decision, documented like any other vendor decision in the practice.
The technology earns its place when it is treated the way we treat a new clinician in the practice: useful from the first week, trusted gradually, and reviewed until the pattern of its mistakes is known.