skip to main content
international guides

AI Scribes in the NHS: What They Are (and Aren’t)

A non-hype guide to ambient scribing: safety, accountability, data flows, and why governance matters more than features.

Review date
Jan 2026
Reading time
9 min read
Author
Dr Kola Tytler MBBS CertHE MBA MSt MRCGP
On this page
IMG pathwayThis guide is updated for 2026 regulations (UK/US/Aus/Can). Always verify with official bodies.

Key points

  • Treat ambient scribing as clinical documentation support, not autonomous decision-making.
  • Assume the clinician remains accountable for the record; review and edit is non-negotiable.
  • In the NHS, compliance expectations typically touch DTAC, clinical safety (DCB0129/0160), and (if relevant) MHRA medical device status.
Ambient scribing products listen to a consultation (or parts of it) and generate structured documentation (e.g., a draft note, letter, or coding suggestions). The product value is administrative acceleration — not clinical reasoning. The practical risk is that teams accidentally treat drafts as ‘truth’, or introduce uncontrolled patient data flows into third-party systems.

The safest operating principle

If it ends up in the clinical record, a clinician must read it, correct it, and consciously sign it off. Any workflow that bypasses that review step is asking for downstream harm.
1

Step 1 — Define the use case tightly

Decide what you want produced: consultation note, referral letter, patient summary, admin task list. Narrow scope reduces risk and makes evaluation possible.
2

Step 2 — Map your data flow

Write down what audio/text leaves your environment, where it is processed, what is stored, and for how long. If you can’t describe the flow, you can’t govern it.
3

Step 3 — Decide what must never be captured

Agree ‘red lines’ locally (e.g., sensitive third-party information, certain categories of patient data, safeguarding details). Train staff to pause/disable capture when needed.
4

Step 4 — Make review explicit in the workflow

Build time for verification (even 30–60 seconds) and define what “good” looks like: completeness, correct negatives, correct meds, correct safety-netting language.
5

Step 5 — Pilot with measurement

Measure time saved AND error rates (near misses, corrections per note, missing key items). If quality drops, the ‘time saved’ is debt.
SourceNHS England: Guidance on AI-enabled ambient scribing products
Open Link
SourceNHS England: DTAC (Digital Technology Assessment Criteria)
Open Link
SourceGMC: Artificial intelligence and innovative technologies (professional standards)
Open Link

Official Sources

NHS England — Guidance on AI-enabled ambient scribing products
NHS England — Digital Technology Assessment Criteria (DTAC)
GMC — AI and innovative technologies (how GMC guidance applies)
Your learning path

Save your place and choose what comes next

sign in to save your learning
Work through this guide (5 steps)

Use the actions from this article to track what you have worked through.

Sign in using the reading controls above to save your steps.

  • Decide what you want produced: consultation note, referral letter, patient summary, admin task list. Narrow scope reduces risk and makes evaluation possible.

    Return to this section

  • Write down what audio/text leaves your environment, where it is processed, what is stored, and for how long. If you can’t describe the flow, you can’t govern it.

    Return to this section

  • Agree ‘red lines’ locally (e.g., sensitive third-party information, certain categories of patient data, safeguarding details). Train staff to pause/disable capture when needed.

    Return to this section

  • Build time for verification (even 30–60 seconds) and define what “good” looks like: completeness, correct negatives, correct meds, correct safety-netting language.

    Return to this section

  • Measure time saved AND error rates (near misses, corrections per note, missing key items). If quality drops, the ‘time saved’ is debt.

    Return to this section

United Kingdom · Starting work · Article 7 of 28