Key points
- The clinician who signs an AI-drafted note is responsible for it. Medicare says so directly for AI scribes, and Australia's regulator says the practitioner checks the accuracy of records created with generative AI.12
- In a review of 356 AI-drafted notes by 31 US physicians, omissions were the most common error (18%), then hallucinations (11.5%) and accidental inclusions (9.3%).3
- 94.7% of those 356 notes were free from significant errors, and 5.3% held an error that could cause serious harm if left uncorrected.3
- Review in a fixed order. Check what is missing first, then what was added, then medicines, numbers, sides and the plan.
- wai drafts the note, letters and suggested coding, and only the clinician can sign them.5
Summary
The clinician who signs an AI-drafted note is responsible for it, exactly as for a note they typed themselves. A fixed review of about two minutes, aimed at the errors studies find most often, lets the clinician sign with confidence.
Every note is drafted by the time the patient leaves, and the clinician's work is a two-minute review and a signature. The routine below keeps that review fast and aimed at the errors drafts actually contain.
See it in your clinicThirty minutes on your own patients and doctors.
Request a demoSee a note drafted from a real consultation and signed after review, on a workflow from your own clinic.
| Check | What to look for | Why it comes in this order |
|---|---|---|
| 1. What is missing | Symptoms, history, examination findings or decisions discussed but absent from the note | Omissions were the most common error, in 18% of reviewed notes |
| 2. What was added | Findings, advice or history that were never said in the room | Hallucinations appeared in 11.5% of reviewed notes |
| 3. Medicines and doses | Drug names, strengths, frequency, starts, stops and allergies | A single wrong word here changes treatment |
| 4. Numbers, sides and negatives | Observations, results, left and right, and pertinent negatives that were stated | Easy to transpose and hard to spot on a quick read |
| 5. Who said what | Patient-reported history kept separate from the clinician's findings and opinion | Attribution decides how the note reads in a complaint or claim |
| 6. Plan and safety-netting | Follow-up, referrals, tests ordered and the advice given on when to return | The plan is what the next clinician acts on |
| 7. Irrelevant content | Small talk, third-party details or material from another patient | Accidental inclusions appeared in 9.3% of reviewed notes |
| 8. Letters and codes | Any letter or suggested code drawn from the note, checked after the note is corrected | Downstream documents inherit any error left in the note |
Responsibility stays with the signature
In the US, the Medicare signature fact sheet tells clinicians who use a scribe, including artificial intelligence technology, to sign the entry to authenticate the documents and the care they provided or ordered.1 The scribe, human or software, is not required to sign at all.1
In Australia, the national regulator's guidance on AI says the practitioner remains responsible for delivering safe and quality care, and is responsible for checking the accuracy and relevance of records created using generative AI.2 UK clinicians answer to their own professional regulator, and UK data protection law requires every reasonable step to keep personal data accurate and to correct inaccurate data without delay.4
What AI drafts get wrong, and how often
In a two-month pilot at a US academic health system in 2024, thirty-one physicians across several specialties used an ambient scribe to draft 7,545 notes, and evaluated 356 of them against a structured rubric.3
Omissions led, in 64 of the 356 notes (18%). Hallucinations, meaning content that was not said, appeared in 41 notes (11.5%), and accidental inclusions of irrelevant material in 33 (9.3%). Bias was rare, in 4 notes (1.1%).3
Of the 142 errors found, 119 were rated mild to moderate, and 19 notes (5.3%) contained an error rated as serious or imminent risk if not corrected.3 The authors concluded that careful clinician review remains imperative.
How much clinicians edit
The median note had 9.0% of its words edited, and 14.9% of 960 notes were signed with no edits at all.3 Individual habits ranged from 1.9% to 69.3% of words changed.3
The routine below keeps the read fast enough that every note is checked before it is signed.
A two-minute review before signing
Review in the order the errors occur. Start with what is missing, because omissions are the most common error and the hardest to see in a note that reads well. Then check what was added, then the details where a single wrong word changes care.
Run the table below on every note for the first two weeks, whatever tool drafts it. After that, keep the top four checks for routine visits and the full list for complex ones. The guide to AI scribe accuracy covers how to measure a tool's error rate in your own clinic.
How wai handles the signature
wai Clinicians drafts the visit note during the consultation, then the referral letter, result explanation and suggested coding, and holds each one for the clinician to review and sign.5 It cannot sign anything on the clinician's behalf.5
The clinician edits the draft and signs it, and only then does it become part of the record, with the clinician's attention on the patient throughout the consultation.
Where to start
Write the review routine into your practice policy and print the table for the first fortnight. Agree that no note is signed without the omission check and the medicines check, however short the visit.
After two weeks, pull ten signed notes, compare them with your memory of the consultation or the transcript, and log any error that survived review. If omissions dominate, slow down on the first check. Before switching the scribe on, settle how patients are told, using the consent form and script.
Bring a recorded consultation. We draft the note live and you review it with the routine on this page.
Frequently asked questions
Who is legally responsible for an AI-drafted clinical note?
Does the AI scribe have to sign the note or be named in it?
Under Medicare's signature guidance, no. The clinician signs, and there is no need to document who or what transcribed the entry.1 Local policy or a regulator elsewhere may ask clinics to record that a scribe was used, so check yours.
What is the most common error in AI-drafted notes?
Omission. In a review of 356 notes, 18% left out something that mattered, compared with 11.5% containing hallucinated content.3 So the review starts with what is missing.
How long should reviewing an AI note take?
For a routine visit, a minute or two with a fixed checklist. Complex visits, new medicines and anything with a safeguarding element deserve a full read against the transcript.
Can I sign a note without editing it?
Yes, if it is right. In one pilot, 14.9% of 960 notes were signed unedited.3 What matters is that the note was read and checked before the signature.
Can wai sign notes or letters for me?
No. wai drafts notes, letters and suggested coding for the clinician to check and sign.5
Sources
- 1Centers for Medicare & Medicaid Services, 2025Complying with Medicare Signature Requirements (MLN905364)Read 14 October 2026Read source
- 2Australian Health Practitioner Regulation Agency (Ahpra), 2024Meeting your professional obligations when using Artificial Intelligence in healthcareRead 14 October 2026Read source
- 3JMIR Medical Informatics, Taylor and colleagues, 2026Quality of Clinical Notes Created by Ambient Listening Generative AI, Pragmatic Prospective Pilot StudyRead 14 October 2026Read source
- 4Information Commissioner's OfficePrinciple (d), AccuracyRead 14 October 2026Read source
- 5wai, 2026wai for clinicians, AI medical scribeRead 14 October 2026Read source
Every note drafted by the end of the consultation, ready for your signature
Fifteen minutes with a clinician who uses wai daily, on a consultation of your own. We reply within one working day.
Partner with us
See it on your own patients, then build it with us
Thirty minutes on your own patients, doctors and rules, in your market. We reply within one working day, and the clinics that partner with us early shape what we build next.
See what wai does between the visits