GUIDES · 10 MIN READ

How to choose an AI medical scribe: a checklist for outpatient clinics

Twelve questions to ask on every AI scribe demo: how notes are drafted, what happens to recordings, EHR integration, coding, patient summaries, security and cost.

The ScribeLive team · October 8, 2026

Every AI scribe demo looks good. The visit is staged, the audio is clean and the note appears like magic. The differences that matter show up later: on a noisy Tuesday, in a compliance review, or when your front desk asks why they still need a second system. This checklist is the set of questions we'd want answered before signing with anyone, including us.

1. How is the note actually drafted?

Ask whether the note is drafted live during the visit or after the recording stops, and how long it takes on a typical 15-minute visit. Both approaches can work; what matters is that the clinician can review the draft before the next patient.

Then ask what the AI is allowed to write. A good scribe documents what was said. It should not add a diagnosis, a cause or an exam finding that nobody mentioned. Ask the vendor how they prevent that, and test it: mention a symptom without naming a diagnosis and see what the note says.

2. Can I check any line against the recording?

You sign the note, so you need a fast way to verify it. Look for assessment and plan points linked to the moment they were said, and a transcript that shows who was speaking. If checking a line means scrubbing through a recording, it won't happen.

3. What happens to the recording?

This is the question compliance leads care about most. Get specific answers:

Where is audio stored, and is it encrypted?
When is it deleted: at sign-off, after a fixed period, or never?
What happens to a recording if the note is never signed?
Which third parties (for example the AI model provider) process the audio, and are they covered by a Business Associate Agreement?

At ScribeLive the recording is deleted when the clinician signs the note, and any unsigned recording is deleted automatically after 72 hours by default. Our security page explains the full lifecycle.

4. How does the note get into our EHR?

“Integrates with your EHR” can mean anything from a one-click write-back to copy and paste. Ask exactly how the signed note, codes and any structured fields arrive in your EHR, who sets it up, and what it costs. If the answer is an API, ask whether your EHR vendor or integration partner has used it before.

5. Does it suggest codes, and which ones?

Some scribes suggest ICD-10 diagnosis codes, some suggest billing (CPT / E&M) codes, some both. Ask which, how many per visit, and whether each comes with a confidence score or evidence. The clinician should always have the final say.

6. Will it fit the way we document?

Ask whether you can add clinic-specific fields, such as screening scores, that the AI fills in on every note. And ask what happens when you'd rather not record: can you write or dictate the note in the same place?

7. What does the patient get?

A note helps the clinician. A plain-language summary helps the patient follow the plan, and saves the phone calls that come when they forget it. Ask whether the scribe produces one, and how it reaches the patient. We wrote more about this in plain-language visit summaries.

8. Does it help the rest of the clinic?

Documentation is one part of the visit. If your front desk runs a separate queue system and your clinical leads track protocols in PDFs, ask whether the scribe can replace any of that. Fewer systems means fewer logins and fewer handoffs.

9. How are clinics kept separate?

For multi-site groups and for any cloud product, ask how one clinic's records are isolated from another's, who inside your organization can see drafts, and whether every view of patient data is logged.

10. What does it run on?

Does it need an app install, specific hardware or a particular browser? Can a doctor start on a phone today, or is there an IT project first?

11. How is it priced, and what happens when we hit a limit?

Per provider, per visit or per AI session? Ask what happens at the end of the allowance: does the scribe stop, or does a bill arrive? Predictable beats cheap.

12. What doesn't it do?

The best vendors will tell you. Ask directly, and write the answers down. Here is our own list.

Run the demo on your own visits

A staged demo shows you the happy path. Ask to try the scribe on real visits from your clinic, with your accents, your specialties and your background noise, and judge it on the notes you'd actually sign.

Frequently asked questions

What should I ask an AI scribe vendor about HIPAA?

Ask whether they sign a Business Associate Agreement, which third parties process the audio and whether those are covered by a BAA too, where data is stored, whether it is encrypted, when recordings are deleted, and whether every view of patient data is logged.

Should an AI scribe suggest diagnosis codes?

Code suggestions save time if they come with a confidence score or evidence and the clinician confirms every code before signing. Ask whether the vendor suggests ICD-10 diagnosis codes, billing codes or both.

How do I test an AI scribe fairly?

Try it on real visits from your own clinic rather than a staged demo, and judge it on whether you would sign the notes it drafts with only light edits.