The paperwork crisis behind the AI scribe revolution

Before understanding why AI scribes have achieved remarkable penetration in NHS primary care, it helps to understand the problem they are solving.

UK GPs operate in a system under extraordinary administrative pressure. A typical GP appointment lasts 10–15 minutes. A significant portion of that time — in many practices, the majority — is spent on documentation: writing clinical notes, coding diagnoses, completing referral letters, and updating records. Post-appointment, GPs routinely continue administrative work into evenings — a phenomenon known as “pyjama time” that has become a symbol of the profession's workload crisis.

The cognitive burden is not just about time — it is about quality. When a GP is simultaneously conducting a clinical conversation and trying to document it accurately, both tasks suffer. The patient gets a less attentive clinician; the record gets less complete documentation.

AI medical scribes address this by listening to the appointment conversation and automatically generating structured clinical notes, freeing the clinician to focus entirely on the patient.

The Heidi Health numbers

Heidi Health has become the dominant AI scribe in UK primary care with remarkable speed. As of early 2026:

The Modality Partnership — described as the NHS's largest GP super-partnership, encompassing 360+ GPs across 53 sites — rolled out Heidi network-wide. The reported outcomes from that deployment: a 51% drop in documentation time during appointments and a 61% drop in after-hours administrative work. (Heidi Health)

To translate the 61% figure into human terms: if a GP previously spent two hours after clinic on administrative tasks, Heidi reduces that to 47 minutes. Across a working year, this represents hundreds of hours returned to the clinician — for rest, for additional patient contact, or simply for the sustainability of a career that is at significant burnout risk.

The wider scribe ecosystem: Tortus and Accurx

Heidi is not the only AI scribe gaining NHS traction.

Tortus AI, deployed through X-on Health across 3,500+ practices, was evaluated in a study published by Great Ormond Street Hospital. The study found a 23.5% increase in direct patient interaction time — a direct measure of clinical quality improvement, not just efficiency. (HTN)

Accurx Scribe, integrated into a platform that reaches 98% of GP practices in England, delivers 97% clinical accuracy with reported 35–40% efficiency gains. (iatroX) Accurx's reach through its existing platform means that Scribe is available to almost every GP practice in England without a separate procurement process — a significant distribution advantage.

What GPs actually think: the RCGP/Nuffield Trust survey

Despite Heidi Health's market penetration, the RCGP/Nuffield Trust survey published in December 2025 — covering 2,108 GPs and representing the most comprehensive dataset on AI adoption in UK primary care — reveals a more complex picture.

Only 28% of GPs are using AI tools — up from 20% in early 2024, but still a minority. (RCGP) Among those who do use AI:

57% use it for clinical documentation

45% use it for CPD (continuing professional development)

44% use it for administrative tasks

The productivity gains are clear in the qualitative responses. One GP described saving “a good 60 minutes of admin time each day.” (RCGP) Another offered a more holistic perspective: “It just helps to enjoy your day a bit more.” (RCGP) These are not trivial observations in a profession with among the highest burnout rates in the NHS.

The barriers: why 72% of GPs still aren't using AI

The adoption barriers are specific and addressable — but they require deliberate action.

Lack of employer encouragement: A striking 85% of GPs say their employer had not encouraged them to use AI. (RCGP) In a profession where individual practice variation is high and organisational structures vary enormously, this reflects the absence of system-level adoption strategies.

Lack of training: 95% of GPs report receiving no professional training on AI tools. (RCGP) Informal peer-to-peer sharing is the dominant adoption pathway — which is both inefficient and uneven.

Medico-legal concerns: 89% of non-AI-using GPs cite professional liability as a concern. (RCGP) If an AI scribe incorrectly transcribes a medication or misses a clinical nuance, and that error persists in the patient record and influences a future clinical decision, who bears responsibility? The GMC's position — that clinicians remain responsible for all decisions when using AI (iatroX) — is clear in principle but creates a liability burden that many GPs are not yet comfortable carrying.

Clinical error anxiety: 83% of non-users cite clinical error risk. (RCGP) One GP described inconsistent performance: “Sometimes it does a marvellous job, some other times it doesn't.” The concern is not that AI is always wrong, but that its failure modes are unpredictable.

Automation bias: GPs explicitly worry about the risk of an “inclination to favour decisions generated by machines” — accepting AI-generated clinical notes without adequate review. (RCGP) This is a legitimate and sophisticated concern that reflects clinical training, not technophobia.

Data privacy: 82% of non-users cite patient data concerns. (RCGP) One GP asked pointedly: “They say the data is deleted in 30 days. But how do you know?” (RCGP)

The regulatory framework: CQC, MHRA, and the approved list

Regulation of AI scribes in NHS primary care has developed rapidly through 2025–2026.

CQC Mythbuster 109 (July 2025) established the key consent and governance principles: (HTN)

NHS England's National Chief Clinical Information Officer instructed providers in June 2025 to cease using non-approved AI scribes pending a formal review process. (RCGP) In January 2026, a self-certified registry was introduced for approved ambient voice technology suppliers — providing GPs with a clearer framework for which tools have been assessed as appropriate.

The MHRA launched its AI Airlock regulatory sandbox in Phase 2 (running until March 2026) (Innovation) and a National Commission on AI Regulation in Healthcare (September 2025), with a new regulatory framework expected in 2026.

Patient sentiment: the missing variable

The adoption picture must also account for patient attitudes, which are more cautious than the efficiency data might suggest. CQC research found that 47% of GP patients have negative feelings toward AI in healthcare, versus just 35% with positive feelings. (RCGP)

Practices implementing AI scribes should take patient communication seriously. The CQC's requirement for patient notification is a floor, not a ceiling. Clear explanation of what the AI does, what data it processes, and how it is stored is likely to reduce resistance and build the trust that makes implementation sustainable.

Key statistics at a glance

85% of GPs have not been encouraged by their employer to use AI (RCGP)

95% of GPs have received no professional AI training (RCGP)

47% of patients have negative feelings toward AI in healthcare (RCGP)

MarGen helps healthcare technology companies build the kind of content that gets cited by AI systems and trusted by professionals. Talk to us.