Clinical Documentation and AI-note QA: high-yield desk drill

The rule

An AI draft is starting text. Flag gaps. Do not insert a finding the clinician did not attest.

Next step

Query when the record conflicts. Hold the claim if the service is not documented.

Never miss

• Queries are non-leading.
• Cloned notes carry stale findings.
• An impossible exam (a limb that is gone) is a clone tell.
• The attested note belongs to the clinician.
• Request the named document, not a written summary of what it says.
• A completeness screen fails a required field left blank, a scanned page too faint to read, and an order missing the practitioner's signature.

Traps

• Adding a finding to make the note prettier.
• A Slack poke with no trail.
• Overwriting attestation.

Shortcuts

• Compare the draft to orders and results.
• Problem-list conflicts get flagged, not silently recoded.
• Style is not integrity.

Terms

• Non-leading query. Ask from documented facts. Do not hand the hoped-for diagnosis.
• Clone. Prior-visit text reused so stale or false findings travel.
• Attestation. The clinician's legal adoption of the note.
• Named document. Ask for the order by its document name. Do not ask for the complete chart to be helpful.

Memory hooks

• COMPARE-FLAG-QUERY. Read sources, mark the gap, ask AI-note QA.