Skip to content

The floor

AI can draft a note. It cannot close the chart.

By Elena Voss

Chart-review lead

DocumentationJune 2, 20265 min

A hospital corridor after the visit — the chart still has to close
← All notes

Ambient scribes are good at turning speech into paragraphs. They are not good at knowing which paragraph is missing. The standard-of-care trial, the laterality, the classification of the product, the time statement — those are absences. A fluent draft can hide them.

Coding assignment has the same limit. A model can suggest a code. It cannot sit with the LCD, the procedure note, and the billed units and say they do not reconcile. That disagreement is the job. It is also the part that prevents a denial from becoming a pattern.

We use people who have already failed these charts in training, on purpose, in one specialty. Wound care fails differently from podiatry. Vascular fails differently from dermatology. A general model trained on “clinical notes” does not know those differences. A specialist who works one of them does.

If the tool drafts, someone still has to read the chart the way an auditor will. That someone is the QA layer. Without it, you have a faster note and the same exposure. With it, the draft is a start. The chart is finished.