AI medical scribe
Also known as: AI scribe, ambient clinical documentation, ambient AI scribe, ambient scribe, AI clinical scribe
An AI medical scribe, also called ambient clinical documentation, listens to the conversation between a clinician and patient, with consent, and drafts the visit note automatically using speech recognition and large language models.
Doctors review and sign the note instead of typing it. While AI scribes have quickly become one of the most popular uses of AI in healthcare because they give clinicians time back, they can also produce confident errors, so human review isn't optional.
AI medical scribe, explained at three levels
At home
Think of a court stenographer who also writes the summary. An AI medical scribe listens to your appointment, with your permission, turns the conversation into text, and then drafts the visit note in the format doctors use: your symptoms, the exam findings, and the plan. Your doctor reads and corrects the draft before signing it, so they can face you during the visit instead of a keyboard.
At work
An AI medical scribe is software that records a clinical conversation, transcribes it, and uses a large language model to draft a structured note for the clinician to review. For health organizations, it targets documentation burden, a leading cause of burnout. The business case rests on time saved per visit, less after-hours charting, and clinician retention, balanced against licence costs and the need for careful review.
At home
An AI scribe runs a pipeline. First, speech recognition turns audio into text. Next, a step called diarization works out who said what, separating the doctor's voice from yours. Then a large language model reads the transcript and rewrites it into note sections, such as history, exam, assessment, and plan, leaving out small talk. The doctor reviews the draft. Depending on the product, the audio recording is deleted once the note is finished.
At work
Ambient scribes capture audio on a phone or room device, stream it to a vendor cloud for transcription, then prompt a large language model with specialty templates to produce a note. Better products integrate with the EMR, writing drafts directly into the chart rather than requiring copy and paste. Deployment needs patient consent workflows, a privacy impact assessment, and reviewer training. Typical errors are omissions, statements attributed to the wrong speaker, and hallucinated findings that were never discussed.
At home
Language models generate fluent text by predicting likely words, and that fluency hides errors. A misheard drug name or an invented exam finding appears in the same confident prose as everything else, so a tired clinician skimming at the end of the day can miss it. While AI scribes clearly save time, in my opinion the real risk is not that they are bad; it is that they are good enough that people stop reading carefully.
At work
Accuracy varies with accents, languages, background noise, and specialty vocabulary, and Ontario's Auditor General reported in 2026 that every approved scribe product tested had at least one inaccuracy. Evaluate data residency, since many provinces expect Canadian hosting; audio retention; whether vendors train models on your data; and liability for errors in signed notes. Measure documentation time, note-quality audits, and patient experience, not just adoption, and repeat audits after vendor model updates.
Real-world examples
At home
Sophie noticed her family doctor in Kitchener asked permission to use an AI scribe at her appointment. The doctor maintained eye contact the whole visit and later fixed one small error in the note before signing it.
At work
Thomas is a family physician in rural Saskatchewan who used to spend two evenings a week finishing charts. After enrolling in a funded AI scribe program, he finishes most notes the same day, though he still corrects medication names the tool mishears.
How it compares
| feature | AI medical scribe | Human scribe | Clinician typing |
|---|---|---|---|
| Who writes the note | AI drafts, clinician reviews | Trained person drafts | Clinician |
| Cost | Subscription | Salary | Clinician time |
| Error type | Confident hallucinations | Omissions or misunderstandings | Shortcuts and copy-paste |
| Scalability | High | Limited | Not applicable |
AI scribes scale documentation help, but shift the clinician's job to careful review.
For analysts
Before rolling out an AI scribe, I'd set up regular note-quality audits alongside time-savings tracking. Speed gains are easy to measure; subtle errors only show up if someone looks for them.
History & origin
Human medical scribes became common in US emergency departments in the 2000s and 2010s as electronic records increased documentation burdens. Nuance, a speech recognition company, launched its Dragon Ambient eXperience (DAX), and Microsoft acquired Nuance in 2022 for about US$19.7 billion. Large language models made fully automated drafts practical in 2023, and startups such as Abridge, Ambience, and many Canadian companies entered the market. In 2025, Canada Health Infoway launched a national AI scribe program funding licences for primary care clinicians; by February 2026, more than 12,000 clinicians had enrolled. Ontario's evaluations of scribe products found that all tested tools sometimes produced incorrect or incomplete information. In my view, AI scribes are the right first job for AI in medicine: high value, with a clinician checking every output.