An ambient scribe listens to the consultation and writes the note, the letters, and the codes — the clinician reviews and files. The technology is converging fast; the market isn't being won on transcription quality. It's being won on distribution into existing clinical workflow: one player rides a messaging platform already in 98% of England's GP practices, one rides the practice phone system, and one rides the individual clinician's pocket. The UK is the cleanest live experiment in how AI actually enters a regulated profession — and the lesson is not about models.
An ambient AI scribe captures the audio of a consultation — face-to-face or telephone — and generates the structured clinical note, referral and patient letters, clinical coding, and follow-up tasks. The clinician reviews, edits, and files to the patient record. The consistent early evidence claim across vendors is minutes saved per consultation and, less measurably, a clinician who looks at the patient instead of the keyboard.
Why it matters beyond convenience: documentation is the highest-value daily habit in clinical software. Whoever owns the moment of writing the note owns the workflow — which is why this market is the flanking attack on legacy practice management, not a feature of it.
| Player | The rail | Verified shape |
|---|---|---|
Accurx Scribe(engine: Tandem Health) | The comms platform | Accurx is used by 98% of GP practices in England and staff in 70% of NHS trusts; Scribe launched April 2025 to 200,000+ NHS staff on that installed base, writing back to EMIS and SystmOne. By March 2026: supplying a large-scale NHS hospital rollout, plus a four-year multi-trust agreement covering 10,000 clinicians. |
Tortus(Surgery Intellect) | The phone system | Strategic partnership with X-on Health (May 2025) embeds the scribe in Surgery Connect — the cloud telephony used by 3,500+ UK GP practices handling 40M+ calls a month — so it transcribes phone and front-desk conversations natively. MHRA Class I registered, Class IIa pending; early users report ~4 minutes saved per consultation. |
Heidi | The clinician's pocket | Free tier for individual clinicians, then practice and enterprise plans; a clip-on wearable mic (Heidi Remote) that follows the doctor rather than the desk. States it powers "the largest AI scribe procurement in NHS history" — 70,000 clinicians, 15 NHS trusts, 1,200+ GP practices — and lists dental group Portman Dentex among its clients. |
Numbers are vendor- or partner-stated and dated here deliberately. Adoption figures in this market are announcements, not audits — useful for shape and momentum, not for precision. All figures above are as published April 2025 – March 2026.
None of the three leaders is winning on a proprietary model — transcription and summarisation quality are converging across the category. What differs is the path into the workflow. Accurx didn't build a better scribe; it attached a scribe to a platform clinicians already had open. Tortus put the scribe inside the phone call that was already being handled. Heidi bypassed the institution entirely and recruited the individual clinician — free tier first, hardware next, procurement later.
This is the same finding the tree keeps hitting in legal AI and elsewhere: capability is becoming undifferentiated; distribution and workflow position are not. For anyone building or buying in this category, the question isn't "whose model is best" — it's "who is already inside the moment where the work happens."
UK scribes sit under MHRA medical-device rules — Tortus states Class I registered with Class IIa pending. Classification determines the evidence burden; ask any vendor for theirs.
DTAC, clinical-safety standard DCB0129, the Data Security & Protection Toolkit, Cyber Essentials Plus — the checklist that decides who can sell into the NHS at all. It's also a useful proxy checklist outside the UK.
Every serious player files nothing without clinician review. The scribe drafts; the clinician remains the author of record. Any deployment that drifts from this has changed risk category.
An omitted negative ("no chest pain") or a fabricated detail in a clinical note is not a typo — it's a patient-safety event. Review discipline is the control, and it erodes exactly as trust grows.
Recording consultations requires patient consent processes, retention rules for audio, and clarity on where processing happens — which is where the SA section picks up.
A consultation recording is special personal information under POPIA — health information, processed with explicit consent, with the section 72 cross-border test applying the moment a cloud scribe processes audio offshore. An SA practice adopting any of these tools needs the consent script in the room, the retention policy for audio, and a straight answer from the vendor on where transcription and summarisation actually run. That's the same residency discipline the tree applies to every AI workload — the stakes are just higher here.
The distribution lesson transfers; the rail doesn't exist yet. South Africa has no Accurx — no single platform inside nearly every practice. The closest equivalents to a rail are the medical-aid claims switches, the incumbent practice systems, and, in reality, WhatsApp-first patient communication. Which means the UK's endgame — the scribe arriving through something you already use — hasn't been built here. For an SA builder, that's the actual opportunity this leaf points at: a POPIA-compliant scribe wired into local billing and communication rails, before an international player localises.
Vendor and partner announcements, plus independent trade coverage. Last reviewed 2026-08-12.