An AI scribe for medical notes listens to the visit, transcribes it, and drafts a structured note you review and sign. Under the hood it runs a short pipeline: ambient capture, then speech-to-text, then a language model that shapes the transcript into a SOAP note. The better tools add ICD-10 suggestions and a prescription draft. What each tool lists on its pricing page is the ceiling, so read it literally.
That’s the mechanism. The part that actually decides whether a tool earns a place in your clinic is what happens after the draft appears, so most of this page is about the six things worth checking before you sign.
Key takeaways
- An AI scribe runs four steps: capture, transcribe, structure into a SOAP note, then you review and sign. The draft lands about 2 minutes after the visit.
- Beyond the note, good tools return ICD-10 suggestions and a prescription draft, all draft-only until a clinician signs.
- Primary-care physicians log a median of 36.2 minutes of EHR time per 30-minute visit, which is the burden these tools target.
- Six criteria separate the field: note quality, edit burden, EHR export, language coverage, security and BAA, and price.
to a review-ready draft note after the visit ends
returned with the note, alongside a prescription draft
visit audio processed in memory, discarded at note draft
How does an AI scribe turn a visit into a note?
Four steps, and all of them happen during or right after the appointment.
It listens first. You open a session on your phone or laptop, get the patient’s consent, and run the visit like you always do. The scribe captures the conversation ambiently. Nobody dictates.
Then it transcribes. Speech becomes text in near real time. The solid systems handle two speakers, interruptions, and the way a real consult loops back on itself. The hard case is a full room with a relative answering half the questions in an accent the model hasn’t heard much of.
Next it structures the transcript. A language model reads the text and sorts it into subjective, objective, assessment, and plan. This is where products split. Some stop at the note. Others draft the pieces around it too.
Then you read it and sign. The draft shows up about two minutes after the visit ends. You fix what’s wrong, add what the room didn’t say out loud, and sign. Nothing reaches the chart without you.
Here’s the canonical version of that flow, the way our tool does it. AI Medical Scribe is one module inside Practice Copilot, the platform Patient Square builds. It listens during the visit and hands back a structured SOAP note, ICD-10 suggestions, and a prescription draft, ready to review and sign about two minutes after the visit. If you want to watch it run on a real visit type, book a short demo and bring your own note format.
What does the AI actually hand back?
Baseline for any credible product is one thing: a structured note. Past that, read the feature list closely, because “scribe” gets stretched to cover a lot.
Some tools give you the SOAP note and stop there. Others draft the downstream artifacts. AI Medical Scribe by Patient Square returns ICD-10 code suggestions and a prescription draft with the note. Two things to be clear about, because vendors blur them. Those ICD-10 codes are suggestions for you to confirm, not a coding engine that finalizes billing. The prescription is a draft for you to review, not an order that transmits to a pharmacy. The clinician stays the author of every one.
That draft-only line is the honest one. A tool that claims to code the chart or send the script by itself is either overselling or building something you shouldn’t want. The point of the review step is that a draft a model wrote is still a draft a model wrote.
The six criteria that actually separate AI scribes
Every vendor demo looks good on a quiet, scripted visit. The differences show up on a 14-patient Tuesday. These are the six things worth pressure-testing before you commit, and the honest questions to ask about each.
| Capability | What to check | AI Medical Scribe by Patient Square |
|---|---|---|
| Note quality on a real visit | Structured SOAP note, ~2 min after the visit | – |
| Edit burden after the draft | Review-and-sign; you correct, not rewrite | – |
| EHR export format | EHR-ready export: PDF, HL7, FHIR, every plan | – |
| Language coverage | English and more languages; note in clean clinical English | – |
| Security and BAA | Aligned with HIPAA Security Rule; BAA available; audio not stored | – |
| Price, published | From $79/clinician/mo, annual Assist plan; 7-day trial | – |
Note quality on a real visit
Edit burden after the draft
EHR export format
Language coverage
Security and BAA
Price, published
Note quality is first because it’s the whole job. A clean draft on a single-speaker consult is table stakes. Ask to see the tool run on your messiest visit type, not the vendor’s tidy demo script. Read a real note it produced end to end.
Edit burden is the criterion buyers skip and regret. A tool that drafts a note you then rewrite from scratch hasn’t saved you anything. The goal is review-and-sign, where you correct a line or two and move on. Time yourself editing three real drafts during a trial.
EHR export decides whether the note gets where it needs to go. Some vendors write into named EHRs; others hand you a file. We offer EHR-ready export in PDF, HL7, and FHIR on every plan, so the note leaves in a format your record system accepts. Note that’s an export, not a certified two-way interface. Ask each vendor exactly what “integrates with your EHR” means, in one sentence.
Language coverage is where “multilingual” marketing goes to die. Read the language list literally, then test it on your own patients. We capture English and more languages and always return the note in clean clinical English.
Security and BAA is non-negotiable for PHI. The floor is a signed BAA plus encryption in transit and at rest. We map our safeguards to the HIPAA Security Rule, offer a BAA to every US customer, and don’t store the audio: it’s processed in memory and discarded once the note is drafted. Our full posture is on the security page. No vendor is “HIPAA certified,” because no such certificate exists; treat that phrase as a red flag.
Price should be on a page, not behind a sales call. Published US scribe pricing runs roughly $39 to $199 per clinician per month, and several big names publish nothing. We think a per-clinician software subscription that hides its price tells you how the renewal will go. Ours starts at $79 per clinician per month on the annual Assist plan, with unlimited visits and notes.
For a deeper version of this checklist, our guide on how to evaluate an AI medical scribe walks through the trial itself. And SOAP note quality covers what a good draft should and shouldn’t get right.
When is a human scribe or a template the better fit?
We’ll be straight about this, because a scribe isn’t the answer for every visit.
A human scribe still wins in a few cases. If the visit is chaotic enough that a person needs to interpret it live, a trauma bay or a packed OPD with three people talking, a trained human in the room reads the situation in a way software doesn’t yet. If your patients won’t consent to any recording, a human writing from observation may be the only path. The trade is cost: a human scribe is a salary plus hiring and turnover, and it scales one clinician at a time.
A template-only tool can be the better fit too. If your notes are short and repetitive, say a dermatology follow-up you already document in 40 seconds with a macro, an ambient scribe is solving a problem you don’t have. Smart templates and quick-text are cheaper and faster for that pattern. The math flips the moment your notes are long, variable, and piling up after clinic. Primary-care physicians log a median of 36.2 minutes of EHR time per 30-minute visit, per a 2023 JAMA Network Open study; that’s the exact burden an ambient scribe targets, and a template doesn’t.
So the honest verdict: pick the human scribe for the hardest, non-recordable rooms, pick the template for short repetitive notes you already write fast, and pick an AI scribe when the after-hours charting is the real problem. If that last one is you, see what’s on each plan before you shortlist anything.
What the AI scribe still can’t do for you
Worth saying plainly, since honest limits are how you judge a vendor.
Drafts contain errors. Models mishear medication names, occasionally merge two complaints, sometimes write something plausible that didn’t happen. The review step is load-bearing for exactly that reason. Distrust any pitch that implies you can skip reading the note.
Noisy rooms are still the hard case. A quiet consult is easy; a crowded clinic with a caregiver answering half the questions is where transcription quality varies most across products. Test your real patient mix, not the demo’s.
And the chart stays yours. No scribe, software or human, moves the documentation responsibility off the signing clinician. The ICD-10 codes are suggestions you confirm. The prescription is a draft you decide on. A good tool makes those decisions faster and better documented. It doesn’t make them for you, and you wouldn’t want a tool that did. If you want to see where those lines sit in practice, our audio-retention explainer covers what happens to the recording, and the pricing page lists exactly what ships on each plan today.
Frequently asked questions
The FAQ above covers the recurring questions: how the note gets written, whether the draft is accurate enough to trust, whether it reaches your EHR, what else it hands back, what happens to the audio, when a human or template fits better, and what it costs. For the full plan breakdown, see the pricing page, and a demo shows the flow on your own visit types.