Ambient AI scribes listen to your consultation and write the clinical note for you. They have moved from novelty to daily tool in a remarkably short time, and until now nobody had good UK numbers on who is using them and what they actually think. A new study in BMJ Health & Care Informatics changes that: a nationwide survey of 1,003 UK GPs, the largest of its kind to date.
The headline: adoption is early but accelerating fast, the time savings are real, and so are the errors; especially in exactly the consultations where accuracy matters most.
Key takeaways
- 14% of surveyed UK GPs already use an ambient AI scribe; another 39% intend to adopt one soon
- 80% of users say documentation time fell; 70% report a lighter cognitive load
- Errors are common: 44% of users find mistakes in 10 to 30% of their AI generated notes
- Error risk climbs in multiparty, complex and non English consultations
- 37% of users do not routinely ask patients for consent
- You remain legally responsible for the record, not the software
What the study looked at
Researchers led by Charlotte Blease surveyed UK GPs through Doctors.net.uk in August 2025, asking about use of ambient AI scribes: background systems that capture the consultation dialogue and generate documentation automatically. Of 1,003 respondents, 141 were current users, and the study drilled into their experience of errors, workflow and consent. The authors are careful to call this an early adopter snapshot rather than a population estimate, since respondents were a convenience sample of digitally engaged GPs.
One product dominated: 86% of users ran Heidi Health. Adoption skewed towards GP partners and principals and larger practices, and away from locums and GPs aged 56 and over; a pattern the authors suggest reflects who has the autonomy and infrastructure to introduce new tools.
The good news: less typing, lighter heads
The efficiency findings are striking for a profession drowning in documentation:
- 80% of users reported spending less time on documentation; 30% said the reduction was large
- 70% reported reduced cognitive load during consultations
- 55% rated the AI generated notes as better than standard documentation, needing only minimal revision
The free text comments echoed the numbers: GPs valued being able to look at the patient instead of the screen. If your working day ends with an hour of unfinished notes, it is not hard to see the appeal.
The catch: errors are routine, and they cluster where it hurts
The same GPs who praised the tools also reported correcting them constantly:
- 44% find errors in 10 to 30% of AI generated documents
- 41% correct nearly all generated documentation in some way
- 32% encounter errors often or always, and 14% have seen errors with significant to critical implications for care
Most errors were judged minor. But their distribution should give every practice pause. Error rates were highest with multiple people in the room (38%), complex medical histories (35%), consultations in languages other than English (31%), speech impairment (21%) and noisy rooms (20%). The authors warn this pattern could feed the inverse care law: the patients who already face communication barriers get the least reliable records.
The practical reading, in the authors’ words, is that this is assisted documentation, not replaced documentation. The time saving comes from editing a structured draft rather than composing from scratch. The moment you stop reading the draft is the moment the risk starts.
Consent: the finding with immediate implications
Only 63% of users routinely sought patient consent before using the scribe. Put the other way, 37% did not; a figure the authors flag as having immediate clinical and regulatory implications. Among GPs who did ask, patients rarely minded: 89% said one in ten patients or fewer declined.
UK guidance on whether consent is strictly required remains unsettled, but the ethics are not: patients can only decline something they know about, and the study found asking costs almost nothing. Australia has already gone further, with its professional regulator writing informed consent for scribe use into the core code of conduct.
The regulatory ground is moving
NHS England issued guidance on ambient scribing products in April 2025, requiring a named clinical safety officer, a data protection impact assessment, proper system integration, compliance with data security and medical device regulation, staff training and ongoing monitoring. By June 2025 its chief clinical information officer was warning organisations to stop using tools not registered as a class I medical device with the MHRA.
One point is not moving at all: responsibility for the accuracy of the record stays with the clinician. Using an AI scribe changes how the note gets written; it does not change whose name is accountable for it.
What this means for your practice
If you are using or considering an ambient AI scribe, this study suggests a short checklist:
- Treat every AI note as a draft. Read it before you file it, every time.
- Raise your guard in the known danger zones: interpreters, several voices in the room, long complex histories, noisy rooms. Consider not using the tool at all in some of them.
- Ask patients, every consultation. Most will say yes; the asking protects the relationship and you.
- Check your tool’s regulatory status against the NHS England requirements, and know who your clinical safety officer is.
- Keep an eye on accuracy for specific patient groups, not just overall performance.
Make your AI scribe journey count towards your appraisal
Introducing an ambient AI scribe is a textbook quality improvement activity, and the reflections almost write themselves: what changed in your documentation time, what errors you caught and what you did about them, how patients responded when you asked for consent. If a scribe error ever reaches a patient record with consequences, that belongs in your significant events log with the lessons learned.
If you use gptools for your appraisal toolkit, a few minutes logging that experience as a QIA entry, with this study as your supporting reading, turns your everyday adoption of new technology into strong appraisal evidence. Every doctor’s first year is free, so there is no barrier to starting your log today.
FAQ
What is an ambient AI scribe?
A system that listens to the clinician and patient talking during a consultation and automatically drafts the clinical note, so the clinician edits a draft instead of typing from scratch.
How many UK GPs use AI scribes?
In this August 2025 survey of 1,003 UK GPs, 14% were current users and a further 39% intended to adopt one soon. Because the sample was self selecting, treat these as an early adopter snapshot rather than a national prevalence figure.
Are AI scribes accurate enough for general practice?
Users report the notes are often good, and 55% rate them better than standard documentation. But 44% find errors in 10 to 30% of documents, and error rates rise sharply in multiparty, complex and non English consultations. The safe pattern is to treat every output as a draft requiring verification.
Do I need patient consent to use an AI scribe?
UK requirements are still being clarified, but NHS England guidance places significant governance duties on users, and the ethical case for telling patients is strong. In this study most patients consented when asked; 37% of GP users were not routinely asking.
Who is liable if the AI scribe gets it wrong?
You are. In both the UK and Australia, responsibility for the accuracy of the clinical record stays with the clinician regardless of what tool drafted it.
Source: Blease C, Kharko A, Garcia Sanchez C, et al. Ambient AI in primary care: an exploratory mixed methods survey of UK general practitioners. BMJ Health & Care Informatics 2026;33:e101847. Read the full study.