Blog

The Pre-Signing Routine: Five Checks Before You Sign an AI-Scribed Note

Published 2026-11-09 · Last updated 2026-11-09

Changelog
  • 2026-11-09: Published.

From October 2025 to October 2026, Centerstone generated about 510,000 notes with ambient documentation and saved roughly 85,000 hours of clinician time, according to data the organization shared with Becker's Behavioral Health [1]. That is the scale AI scribes have reached in behavioral health. At that scale, the review before signature decides whether the time saved stays saved, or is spent later correcting the record. This post sets out a five-check routine to run before you sign, built from the errors our clinical pilot caught in AI-generated notes from standardized synthetic psychiatric encounters.

Why a routine, and why now

The premise is settled. Under Medicare's rules, the clinician who signs an AI-scribed note affirms that the note documents the care provided, whether a person or a model drafted it. The CMS fact sheet on signature requirements states it plainly: "If you use a scribe, including artificial intelligence technology, sign the entry to authenticate the documents and the care you provided or ordered" [2]. We covered what the signature attests in our earlier post on signing AI-scribed notes. This post is the practical companion to that one: what to do in the minutes before you put your signature on the note.

The reason the routine can be short is that the errors are specific. Our clinical pilot scored notes from six vendors across three synthetic psychiatric encounters. Two vendors were flagged "Unsafe as written" on two of three encounters each: Mentalyc scored 79.0 with two critical flags, and Upheal 63.0 with three. The other four scored 83.2 to 96.0 with zero critical flags. The flagged errors were specific: a wrong drug name, a drug attached to the wrong person, a diagnosis the examination contradicted, and findings that never happened in the session. A routine that checks those exact places catches the errors that matter most, without re-reading every line.

The routine: five checks

Run the checks in this order. Medication errors carry the most direct clinical risk, so they come first. Once it is habit, the routine stays short enough to run before every signature.

  • Check 1: Medication names and identity. Read every drug name in the note against the prescription you actually wrote or continued, including dose and the start, stop, or continue instruction. In our pilot, a Mentalyc note documented Lexapro as citalopram and continued it as citalopram. Lexapro is escitalopram, not citalopram, and the error was scored Incorrect and flagged critical, leaving the note "Unsafe as written." Brand and generic names that sound alike are the trap. Check the name, not just the dose.
  • Check 2: Medication ownership. Confirm whose medication each line belongs to. An Upheal note grouped zolpidem with the patient's own bupropion without attributing the zolpidem to his wife, a critical error. In a Mentalyc note, the wife's sleeping pills were labeled with the invented drug name "Collided" when the audio named zolpidem, or Ambien. Household members' medications are routinely discussed in psychiatry sessions, and the note must keep each drug attached to the right person.
  • Check 3: Diagnosis against the examination. Read the diagnosis and the mental status examination together and ask whether they agree. A Mentalyc note diagnosed major depressive disorder with psychotic features while its own mental status examination denied hallucinations. That contradiction was flagged critical. A diagnosis the examination does not support changes the diagnostic record that follows the patient.
  • Check 4: What the note claims you observed. Every examination finding in the note should be something observed or elicited in this session. A Mentalyc note described tearful affect during trauma processing and crying when re-reading an impact statement in session; the crying happened at home, and the invented exam language drew a major deduction. A Freed note invented an abnormal mental status block, including impaired judgment, that was not spoken in the audio. If a finding in the note surprises you, it probably did not happen.
  • Check 5: What the note left out. Omissions do not announce themselves, so check the plan against the decisions and counseling you remember from the session. An Upheal note never documented vitamin D or occasional ibuprofen, both predefined critical items for that encounter. A Mentalyc note advised talking to a primary care clinician about frequent ibuprofen "due to potential interactions" without ever naming bleeding risk. Even the strongest pilot performer left gaps: Heidi scored 96.0 with zero critical flags and still missed risk-related documentation, including an explicit outpatient safety determination in a suicide-risk assessment.

How to run it

Keep the same order every time, so the routine becomes habit rather than a decision. Sign each note per encounter, close to the session, not in an end-of-week batch, when recall of any session is weakest. When a check fails, correct the note before signing. Do not sign and plan to fix it later, because the signature is the affirmation, and a correction after signing may not happen.

If the same failure repeats across notes from one vendor, that is a pattern, not a typo. Raise it with the vendor, test the vendor again on your own encounters, and weigh it when you renew. Documentation scores are a separate question from note accuracy. On our research scorecard, Mentalyc (75/100, #5) and Upheal (71/100, #6) among vendors with a documented BAA; both rankings measure published documentation, not note accuracy. We wrote about why those two kinds of evidence can point in different directions in our post on two rankings that disagree. A vendor can document well and still write notes that need correction, and the routine is how you find out which one you have.

What the pilot does not prove

The pilot is a pilot, and the routine is built on its findings, not on a claim that they generalize. It ran three synthetic encounters, one scored run per vendor per scenario, with text-to-speech audio, and the evaluations were adjudicated by a single psychiatrist scorer, with no second independent psychiatrist review yet. Vendor settings also differed: Mentalyc was tested as an intake note in longer form online, and Upheal on default settings. The flags are a reason to test carefully, not a verdict on every note either vendor will ever produce. The full protocol, including those limits, is published in the pilot methodology. The routine is not a substitute for reading the note. It is a way to read it with attention where errors concentrate, and to make that attention repeatable across a full clinic day.

The bottom line: Centerstone's 510,000 notes show what ambient documentation can save, and our pilot's five error types show where it can still go wrong, so run the same five checks, in the same order, before every signature: drug names, drug ownership, diagnosis against examination, observed findings only, and what was left out.

Sources

  1. Becker's Behavioral Health, "How Centerstone saved 85,000 hours of clinician time with ambient AI," https://www.beckersbehavioralhealth.com/ai-2/how-centerstone-saved-85000-hours-of-clinician-time-with-ambient-ai/, opened 2026-11-08.
  2. Centers for Medicare & Medicaid Services, "Complying with Medicare Signature Requirements" (MLN905364), https://www.cms.gov/mln905364-complying-with-medicare-signature-requirements, opened 2026-11-08.

Who curates this site

Curated by a practicing, board-certified psychiatrist, not named to avoid conflicts. From the site data: 19 vendors tracked, 354 distinct external source pages linked. Last site update 2026-11-20; last targeted vendor verification 2026-11-12; last complete research audit 2026-11-04. How we work · Editorial policy.