Summary
General practice is the hardest test an ambient scribe faces. Ten-minute appointments, five problems in one visit, safety-netting that has to be written down, and a referral letter behind every third consult. A scribe tuned on tidy single-problem specialty clinics meets its match in a Monday morning surgery.
So the buying question for a GP is not which scribe drafts the prettiest note. It is which tool survives the multi-problem consult, writes the safety-netting faithfully, and takes the letters and the admin off your evening rather than leaving them there.
The one randomised trial in this category, run across 14 specialties, cut note-writing time 9.5 percent, about 41 seconds an encounter. Real, and modest. The larger win for a GP sits in everything around the note.
WA\ drafts the note, the referral letters and the codes, and answers the front desk. Built for the whole surgery, not the consult alone.
Book a demoWhat a GP should demand
Four things, in order. The multi-problem consult handled without merging problems into mush. Negations and safety-netting written exactly, because 'no chest pain' becoming 'chest pain' is the category's known failure mode. Referral letters drafted from the consult rather than started from a blank page. And UK data handling you can show an ICO auditor, with UK GDPR processing and DTAC answers on paper.
Any vendor that cannot show you all four on a live consult is selling you a demo, not a tool.
The realistic options
Inside NHS practice, Accurx's scribe is the zero-friction first trial because it lives in a tool you already run, and Nabla carries the category's only randomised evidence. Heidi publishes a usable free tier worth testing. All three are credible, and all three stop at the note.
For private GPs the calculus shifts. Your day includes self-pay billing, WhatsApp enquiries and recalls nobody else will do. WA\ covers the consult, the letters with citations, the coding and the front desk on one record, with a published price and a fourteen day trial.
The safety rule that does not move
Simulated-encounter research puts average scribe hallucination near 7 percent, worst wherever the tool infers rather than records. In general practice that means the examination and the safety-netting. The GP reads and signs every note, whichever product wins.
WA\ shows its working and holds every patient-affecting action for human approval, which makes that reading fast rather than a second job.
The short answer
NHS GP, trial what you already have, then make a specialist beat it on your own consultations. Private GP, buy the system that covers the whole surgery. That is WA\, published price, live this week.
Frequently asked questions
Which AI scribe is best for GPs?
It depends on where you practise. Inside the NHS, trial Accurx's scribe first since it sits in a tool your practice already runs, and look at Nabla for the category's only randomised trial evidence. For private GPs, WA\ covers the consult, the referral letters, the coding and the front desk on one record, is UK-native for UK GDPR, publishes its price, and runs a fourteen day free trial.
Do AI scribes cope with multi-problem GP consultations?
This is the hardest case for every scribe and the first thing to test. Run any candidate on a week of real surgeries and check that separate problems stay separate, negations are written exactly, and safety-netting appears word for word. The category's known failure mode is inferring rather than recording, so the GP signs every note whatever tool is used.
How much time does an AI scribe save a GP?
The only randomised trial, across 238 physicians and about 72,000 encounters, cut note-writing time by 9.5 percent, roughly 41 seconds per encounter, and did not reduce after-hours record time. The larger saving for a GP sits outside the note, in letters, coding and front-desk work, which is the part WA\ covers alongside the consult.
Live this week, not next quarter.
Thirty minutes and you will see the consult, the front desk and the patient record running as one system.