A follow-up note says the wound is smaller. The prior note has three wounds, one measurement set, and no stable identifier. The prose sounds complete, but the comparison cannot be reconstructed.
That is the right kind of failure to put in an AI-scribe trial. Wound documentation is not only a transcription problem. It is a source and continuity problem.
Takeaways
- Let the scribe draft spoken narrative, not generate clinical findings.
- Keep measurements, images, and procedure evidence tied to one stable wound ID.
- Test two visits. A polished note from one encounter does not prove continuity.
- Reject any workflow that turns an omitted fact into a confident sentence.
This is a documentation test, not a wound-assessment, treatment, coding, or coverage protocol. The responsible clinician and the organization’s wound, compliance, billing, and privacy leads define the final record.
Draw a two-record boundary before the demo
Treat the draft note and the supporting clinical evidence as linked records with different jobs.
| Record | It may contain | It must not pretend to do |
|---|---|---|
| spoken-note draft | narrated history, observations, procedure sequence, response, plan | measure, stage, diagnose, select treatment, or invent an omitted fact |
| supporting evidence | approved measurements, images, device output, signed procedure fields | become trustworthy merely because a sentence mentions it |
The boundary matters when a clinician works silently. If the clinician measures length, width, and depth but does not state the values, a microphone cannot recover them. The safe draft leaves the fields absent or visibly incomplete. It does not copy a prior value or convert “looks better” into a measurement.
For some covered wound services, the current CMS contractor article calls for objective measurements and other specific documentation. That article is service and jurisdiction guidance, not a universal wound-note template. Use the rule that applies to the encounter, and ask whether the scribe preserves the clinician’s exact narrated facts without claiming that it establishes coverage (CMS A53296).
Run the two-visit reconstruction test
Start with a new encounter, then create a follow-up for the same wound. Give the reviewer only the two final drafts and the linked evidence. The reviewer should be able to answer five questions:
- Which wound is being compared?
- Which values came from today’s approved measurement process?
- What procedure, if any, did the clinician say was performed?
- Which assessment and plan did the clinician own?
- What is supposed to be checked next, by whom, and when?
A miss is useful. It shows whether the speaking habit, template, or system boundary needs work. It is better to find that miss in a synthetic trial than in a live chart review.
Our wound care documentation template can supply the fields for this exercise. Keep it as a prompt, not an automated clinical decision.
Add cases that tempt the system to guess
The strongest demo is not the smooth visit. Try a follow-up with two nearby wounds, a procedure narrated out of order, and a measurement corrected before the encounter ends. Then ask the vendor to show what happens when the clinician never states a required fact.
Good behavior is plain: the draft reflects the correction, keeps wound identifiers separate, and exposes the gap. Bad behavior is fluent fabrication.
Ask how the team handles photographs too. An ambient note does not automatically need an image. If the clinic uses wound photography, access, consent, retention, and chart linkage should remain inside an approved workflow. Review the vendor’s safeguards and agreement against your own risk analysis; HHS Security Rule guidance is a starting point, not a product endorsement. Patient Square’s current security posture is described on the security page.
Score the draft on provenance, not polish
Use a short scorecard after each case.
| Test | Pass condition |
|---|---|
| wound identity | every observation and procedure maps to the intended wound |
| fact source | dictated values remain distinguishable from linked evidence |
| omissions | missing facts remain missing or visibly incomplete |
| clinician control | assessment, plan, code suggestions, and prescription drafts require review |
| finalization | only the authorized clinician signs the final record |
AI Medical Scribe is part of Patient Square Practice Copilot. It can prepare a structured SOAP-note draft, ICD-10 suggestions, and a prescription draft for clinician review. The suggestions are not final codes, the prescription is not sent, and the scribe does not diagnose or inspect a wound. The clinic still needs a defined review and signing process.
If you are building that process from scratch, use the AI-scribe trial guide for the rollout and the EHR integration checklist for handoff questions. A wound-care trial passes when the record remains traceable even on the awkward cases, not when the cleanest transcript reads well.