The market made "scribe" mean everything
Right now, "AI scribe" can mean:
- a transcript
- a SOAP summary
- templated documentation
- a note generator with zero context
Clinicians hear "scribe" and assume time savings. Leaders hear "scribe" and assume ROI. The gap is that not all scribes produce trustworthy documentation, and not all reduce workload meaningfully.
The four levels of scribe
Level 1. Transcription
You get words. You still do the work.
Level 2. Summarization
You get a note draft. Better, but it can be wrong in subtle ways, and it often misses the longitudinal context that makes a plan correct.
Level 3. Grounded drafting
The system drafts a note and shows sources from the chart. Labs, meds, prior plans. It is harder to hallucinate when the model has to cite.
Level 4. Longitudinal scribe
The system remembers the patient over time and uses that memory to structure the visit:
- what changed since last time
- what is unresolved
- what therapies were tried and why they changed
- what monitoring patterns matter
This is where scribing becomes more than documentation. It becomes clinical leverage.
What clinicians actually want
Clinicians do not want a perfect paragraph. They want:
- fewer clicks
- less retyping
- less hunting
- fewer tasks after the visit
- fewer "I forgot" moments
A scribe that writes faster notes but creates more inbox work is not a win.
The safety test
Ask one question. If the scribe is wrong, how will you know?
Good answers include:
- it cites sources
- it shows what it used
- it highlights uncertainty
- it is deterministic where it should be
- it is auditable
Bad answers include:
- "you just review it"
- "our model is very accurate"
- "we trained on clinician notes"
Healthcare does not run on vibes.
What you should demand from any scribe
A serious scribe product should:
- separate subjective statements from chart facts
- anchor claims to data when possible
- keep a clear edit history
- support structured outputs for downstream workflows
- integrate with routing and tasks, not just notes
If the output stops at the note, the value stops at the note.
The standard to hold
A scribe is not a writing tool. It is part of a clinical system.
If it cannot be grounded, it cannot be trusted. If it cannot be longitudinal, it cannot be great. If it cannot trigger action, it will not reduce workload.