Software & Tools

AI Scribe for Dietitians: What to Check Before You Trust It

Before an AI scribe writes your MNT notes, test it: facts traced to the transcript, units from real time, a valid PES statement, and failures it can't hide.

Disclosure: Farela includes AI notes, so we have a stake here. This checklist applies to any scribe, including ours. For a comparison of specific scribes, see best AI scribes for dietitians.

The note looks perfect. Clean ADIME, a tidy PES statement, a plan with measurable goals. Then you read it slowly. It says the client "eats fast food five times a week." What she said was that she used to, before her diagnosis. It lists 45 minutes of counseling; the call ran 38. And the A1c in the assessment is one she never mentioned.

None of that matters for a wellness check-in. It matters a lot for medical nutrition therapy billed to insurance, because the note is the evidence behind the claim. The minutes support the units, the assessment supports the diagnosis, and a payer who requests your records reads it as your statement, signed by you.

So before you trust an AI scribe with MNT notes, test it on the things that break claims.

Why MNT notes are a harder test than therapy or primary care notes

The checklist

Check Why it matters for MNT How to test it in a trial Red flag
1. Every fact traces to the transcript Invented details become your signed statement Pick 5 sentences; find each one in the transcript A number (weight, A1c, grams) nobody said
2. Tense and negation survive "Used to" and "doesn't" flip the clinical picture Say one past habit and one denial in the mock visit Past behavior written as current
3. Speakers are kept apart A parent or spouse often talks in pediatric and geriatric visits Have a second voice add one fact Family member's words attributed to the client
4. Time comes from the recording Units must match face-to-face minutes Run a 52-minute mock visit Calendar slot used as visit time; charting time counted
5. Units follow the CPT midpoint rule 52 minutes is 3 units; 53 is 4 Check the units at 22/23 and 52/53 minutes Units rounded "to be safe" or missing
6. PES is a real nutrition diagnosis Payers and auditors read it for medical necessity Compare with the PES statement builder "Type 2 diabetes related to..." as the problem
7. Interventions address the billed diagnosis The note has to justify the claim Read the Intervention section against the ICD-10 code Generic advice unrelated to the diagnosis
8. Diagnosis codes are suggestions, not facts A scribe can't know the referral See if it invents codes the referral doesn't support Z68 BMI code listed as primary
9. Failures are loud A silent failed recording means a 9 pm note from memory Cut the connection mid-session No warning; a short note that looks complete
10. Edits and sign-off are tracked Amendments must be identifiable and dated Edit a signed note Changes overwrite the original without a trace

A 20-minute trial protocol

Don't evaluate a scribe on a real client first. Write a short mock visit script with a colleague and plant the traps:

  1. Two numbers: a weight and a meal count. Leave one lab value unspoken.
  2. One past habit ("I used to skip breakfast") and one denial ("I don't drink soda").
  3. A second speaker who adds one fact.
  4. A known length: stop at exactly 52 minutes of face-to-face time.
  5. A referral diagnosis you give the scribe, like E11.9.

Then grade the note against the table above. A scribe that gets 9 or 10 right is a time saver. One that invents a lab value or adds a unit is a liability, however good the prose.

Record the trial the same way you'd record a client: with written consent, and in line with your state's rules. Our recording consent guide has the details.

What to do with the result

If the note is good but stops at the note, you have a scribe. You'll still move the time into units, choose 97802 or 97803, pair the diagnosis and build the claim. That's fine for cash-pay practices and for RDs who like their current EHR's billing. Our AI charting guide covers that setup.

If you bill insurance and the note is only half the work, check whether the tool carries what it measured into the claim. Farela records the visit (video in Farela, in person or by phone), writes the structured MNT note, counts units from the visit time, and drafts the claim with CPT, ICD-10, modifiers and place of service for you to review and submit through your own Claim.MD account.

Where Farela is not the best fit: if you don't bill insurance, a general scribe may be all you need. If you want to stay in your current EHR and only paste notes into it, a standalone scribe is simpler. And if your clients need a portal or app, Farela isn't a patient-facing EMR.

Questions to ask the vendor

The note is only part of the evaluation. Before real clients, get written answers to these:

  1. Where is the audio stored, and for how long? Is it deleted after the note is written, or kept?
  2. Is my data used to train models? If so, can I opt out?
  3. Will you sign a Business Associate Agreement? Don't record a real client before you have one.
  4. What happens when the recording fails? Do I get an alert during the visit or only find out afterwards?
  5. Can I export the note, the transcript and the audit trail if I leave?

Habits that keep any scribe honest

Sources

  1. CMS Medicare Program Integrity Manual, Ch. 3 §3.3.2.4–3.3.2.5 (signatures, amendments, delayed entries)
  2. CMS MLN905364 — Complying with Medicare Signature Requirements
  3. CMS Medicare Claims Processing Manual, Ch. 5, §20.2 (counting timed units)
  4. Academy of Nutrition and Dietetics — Nutrition Care Process
  5. NCPro — Freely available NCP terms
  6. 42 CFR 410.130-410.134 (MNT definitions and referrals)

Sources checked . Payer rules change; verify the member's benefits.

Frequently asked questions

Can a dietitian use an AI scribe for insurance-billed MNT notes?

Yes, as long as you review, correct and sign every note, because the note is your documentation, not the software's. For MNT the note also supports the claim: the documented time supports the units and the assessment supports the diagnosis, so check those parts closely before signing.

How do I know if an AI scribe is making things up?

Test it. Run a mock visit with facts you control, then check five statements in the note against the transcript word for word. Watch for numbers that were never said, a past habit written as current, and something a family member said attributed to the client. A good tool lets you jump from a sentence in the note to the moment in the recording.

Should the AI scribe count my billing units?

It can help if it measures face-to-face time from the recording and applies the CPT midpoint rule (1 unit of 97802 or 97803 from 8 minutes, 2 from 23, 3 from 38, 4 from 53). It should never count charting time or use the calendar slot instead of the real visit. You still confirm the minutes before the claim.

What makes a PES statement from an AI scribe acceptable?

The same things that make yours acceptable: the problem is a nutrition diagnosis you can resolve (not a medical diagnosis), the etiology is a cause your intervention can change, and the signs and symptoms are measurable things you can recheck at follow-up. Reject a PES that restates the referral diagnosis.

Do I need the client's consent to record a session for an AI scribe?

Get it every time, in writing, before recording. Some states require every party to consent to recording, and telehealth adds the client's state to the question. Our recording consent guide lists the rules by state.

Part of Software & workflow. Start with Best Practice Management Software for Dietitians (2026).

Farela

The chart is free. You press Submit.

A free EHR that records the session, writes the note from it, and builds the claim you submit. Billing is 3.9% of what an insurer pays, plus your own Claim.MD plan.

Create free account → No card on file for the EHR