What happens if an AI scribe writes down something the doctor never said, or leaves out something the doctor did? The clinician who signed the note still owns it. In a lawsuit, two versions of the same visit can end up sitting side by side on an exhibit table: the raw audio the software captured, and the polished note that went into the chart.
The distance between what was said and what was recorded is the new fault line in medical malpractice. Experienced medical malpractice attorneys know it changes how a case gets built, how a defense gets argued, and what a patient’s lawyer looks for the moment the records come in.
The scribe hears everything, but the note is a summary
An ambient AI scribe sits in on the visit, listens through a microphone, and drafts a clinical note in the background. The clinician reviews it, edits it, and signs. That workflow sounds clean on paper. In practice, the draft is a probabilistic summary of a conversation, not a transcript of it, and summaries lose things.
A narrative review of ambient scribe performance found an average of 2.9 errors per note across simulated outpatient encounters, with 70% of notes containing at least one error. The overwhelming majority of those errors were omissions, meaning the AI left clinical information out. Some were trivial. Some were not: a patient’s reported loss of appetite that never made it into the record, echocardiogram results that disappeared from a cardiac note.
That is the first thing to understand. The AI usually isn’t inventing wild new facts; it’s dropping real ones.
The audio doesn’t disappear when the note is signed
Here’s the part most patients, and plenty of clinicians, don’t realize. Depending on the software and healthcare provider’s policies, the raw audio and initial AI draft may remain in a vendor’s system after the signed note is filed. Retention policies vary widely, with some systems deleting recordings shortly after processing and others keeping them longer for quality monitoring or other purposes.
If that recording still exists during litigation, it may be discoverable and could become important evidence in a medical malpractice claim. When the recording differs from the final encounter note, even seemingly minor inconsistencies can raise questions about what was discussed, documented, or overlooked during the visit. A family that suspects a missed diagnosis may benefit from requesting preservation of any available recordings and AI-generated drafts early, before routine deletion policies result in their loss.
Automation bias is how a wrong note gets signed
Clinicians are supposed to review every AI-generated note before signing. Most do. But reviewing a 400-word draft that reads fluently is not the same as reconstructing a 15-minute visit from memory, and the pull toward trusting a well-written draft is strong, especially at the end of a long clinic day.
That pull has a name: automation bias. It’s how a dosage transcribed as five milligrams instead of 0.5 milligrams can slip through a review, and it’s how an omitted allergy or medication can end up causing a real adverse event downstream. The signature at the bottom of the note adopts the content and the responsibility that comes with it.
What the discrepancy does to a case
When a plaintiff’s expert can line up the audio against the chart, a few specific patterns tend to drive the case:
- Missing history. The patient described a symptom on the recording, such as chest pain radiating to the jaw or a new-onset headache that woke them up, and it didn’t make it into the note the next clinician read.
- Phantom exam findings. The note describes an exam element that the audio suggests was not performed, or was performed in a way that wouldn’t support the finding as written.
- Edited plans. The audio has the clinician telling the patient one thing (come back in two weeks, go to the ER if symptoms worsen) and the chart documents something looser or something different.
- Consent questions. In states that require all parties to consent to a recording, the mere existence of the audio can open a separate legal front if the patient wasn’t properly notified the scribe was listening.
Any one of those, on its own, isn’t necessarily proof of negligence. Taken together, or paired with a bad outcome, they shift the story. The defense can no longer rest on the clean note alone.
What patients and clinicians should do
Patients should ask whether an AI scribe is being used, and ask what happens to the recording. You’re allowed to decline. If you suspect something went wrong after a visit, request the full medical record promptly and ask specifically whether an ambient recording or AI draft exists for the encounter in question.
Clinicians should read the draft as if a lawyer will one day read it next to the audio, because one day they might. Don’t sign a note you haven’t verified against your own recollection of the visit, and don’t assume the vendor’s retention policy protects you. The chart is yours. So is the distance between it and what the microphone heard.
This content is provided for informational purposes only and is not a substitute for professional advice. AFP editorial staff were not involved in the creation of this content.
Share
