🎧 AI scribes on the ground. Dr Frederick Reynolds, a GP using ambient scribes day-to-day, has written the kind of thoughtful, first-hand review these tools deserve.

📊 The Medicine Central Top 50. Our first of AI tools in UK primary care. One hundred tools longlisted, fifty ranked by deployment, published evidence, and regulatory status. DOWNLOAD THE FULL TOP 50 LIST

PLUS our weekly NICE guidance updates:

💊 QOF 2026/27 adds heart failure four-pillar therapy. The new HF009 indicator (12 points) requires ACEi or ARB, beta-blocker, MRA and SGLT2 inhibitor in HFrEF. Worth auditing your HFrEF list this month.

💷 Semaglutide cardiovascular TA consultation closes Friday. NICE's final draft guidance (expected to publish 28 April) would make around 1.2 million people eligible for weekly semaglutide injections to reduce MACE in established cardiovascular disease with overweight or obesity.

Visit our - on our NICE guidelines page for all updates from the last 2 years (specific for primary care)

By Dr Frederick Reynolds, General Practitioner (MBChB, MRCGP), exploring digital health and innovation products within healthcare

AI scribes, such as Heidi and Tortus AI, are being increasingly used in GP surgeries across the country. They promise to reduce documentation burden by capturing consultations in real time, then automatically transcribing and summarising them into notes that can be integrated into patient records. Having used AI scribes in my day-to-day practice, I can see their potential. The reality is more nuanced.

One of the most meaningful benefits is the ability to focus more fully on the patient. Too often I have found myself typing while a patient is speaking, despite being trained not to do this. AI scribes significantly reduce this need. They allow GPs to be more present in the consultation and engage more naturally with patients.

Emerging data generally supports the use of AI scribes. A study by Modality Partnership, analysing over 2,800 consultations using Heidi, reported a 51% reduction in documentation time and a 61% reduction in after-hours administrative work. Clinicians also reported feeling less stressed and better able to build relationships with patients.¹ Broader systematic reviews echo these findings, although they highlight the need for clinician oversight due to occasional inaccuracies.² ³

In my own experience, I have noticed these inaccuracies. They are often minor, for example a misheard word, but can change the meaning of the whole consultation. Outputs still require careful review. At times this can feel comparable in effort to writing notes from scratch. This raises the question of whether the burden of documentation is actually reduced, or simply redistributed.

I have also found that it takes some adjustment to verbalise examination findings for transcription. It can feel slightly unnatural and can disrupt the flow of the consultation. I suspect this will improve with experience and wider adoption. There are also further practical challenges, notably around capturing and integrating numerical data and coding within the clinical systems.

Integration with existing clinical systems is a key determinant of adoption. It reflects a wider challenge across health technology in the NHS. In practice, the AI scribes I used sat alongside rather than within core clinical systems. Additional steps were still needed to complete routine tasks. Prescriptions, for example, could not be generated within the scribe tools I trialled. I still needed to switch back to the clinical system to prescribe, while the AI-generated consultation note remained separate. The final patient record was not always produced as a single, seamless output. Consultation notes and prescribing activity sat together but were not fully integrated in how they were generated. This made the workflow feel clunky rather than fully embedded in practice. On a minor note, the generated notes did not always match my usual documentation style, so some editing and reformatting was still needed.

It is also worth recognising that writing notes serves an important clinical purpose beyond administration. It supports clinical reasoning, reflection, and decision-making. With AI-generated documentation, the cognitive process shifts towards reviewing and confirming outputs. That has implications for workflow and clinical practice, and it is perhaps one of the more under-discussed changes.

We cannot talk about AI scribes without mentioning governance and data security. GDPR obligations require that patient information is handled accurately, stored securely, and stays fully auditable, particularly when third-party tools are involved. This also extends to medico-legal accountability and clear traceability of how consultation records are generated and edited. NHS England's Ambient Voice Technology Self-Certified Supplier Registry is one step towards a more transparent procurement landscape.

These technologies are evolving quickly, and many of the current limitations are likely to improve over time. There is clear potential to reduce administrative burden and enhance the consultation experience for both patients and clinicians, particularly as integration, accuracy and usability continue to develop. Looking ahead, AI in primary care is beginning to extend beyond documentation, with triage systems, workflow optimisation, and clinical decision support all advancing. AI is set to play an increasingly significant role in general practice, making it essential that these tools are evaluated in real-world practice, with open discussion of their impact and shared learning between clinicians. Their value will depend on how well they integrate into existing clinical systems and workflows, and their adoption should remain thoughtful and appropriately cautious.

References

The market for AI in UK primary care has moved fast over the last eighteen months, and the signal-to-noise ratio has not kept up. Vendors promise transformation. Procurement leads want something they can actually evaluate. Clinicians want to know which tools are real, which are hype, and which are already sitting on their colleagues' desktops.

So we built a market map. We longlisted 100 AI tools currently deployed or in credible pilot across UK primary care. We ranked the top 50 using a composite score:

Ten categories emerged:

1. Tortus AI — Tortus, London, founded 2023. Reaches 3,500+ practices via its partnership with X-On Health. Listed on the NHS AVT Registry. Real-world deployment data growing across multiple ICB pilots. Paid by your ICB or practice.

2. Accurx Scribe — Accurx, London, launched 2024. 3,000 practices and over 1 million consultations summarised. Built directly into the Accurx platform most GPs already use, which removes most of the integration friction other scribes face. NHS AVT Registry listed. Paid as an Accurx module.

3. Dragon Medical One — Microsoft/Nuance, launched 2017. The long-standing global standard for clinical speech recognition, now extended with ambient scribe features. UK deployments run through the AVT Registry. Licensed per clinician.

4. Heidi Health — Heidi, Melbourne, founded 2023. Deployed across Modality Partnership and a growing UK network of 53 sites. Drafts notes, letters and referrals from consultation audio. The source of the 51% documentation reduction figure Freddie cites. Paid by your ICB or practice.

5. Anima Scribe — Anima, London, launched 2024. Built into Anima's triage and consultation platform, it generates notes alongside the online-consultation workflow GPs use for same-day demand. Newer entrant; evidence base still building. Paid as part of the Anima platform.

DOWNLOAD the full ranked Top 50 across all ten categories here:

Nothing final from NICE this week — the Easter gap and the backlog from the new £25,000–£35,000 per QALY threshold (effective 2 April 2026) mean the publication calendar is quiet.

Semaglutide cardiovascular TA — consultation closes Friday 17 April. NICE's final draft guidance recommends weekly semaglutide injection for adults with established cardiovascular disease and BMI ≥27 to reduce MACE, based on a ~20% reduction in major adverse cardiovascular events in the SELECT trial. Around 1.2 million people would be eligible. Final publication expected 28 April. Read the consultation.

Lokelma partial review — final draft guidance released 13 April. NICE is reconsidering sodium zirconium cyclosilicate for hyperkalaemia at a lower potassium threshold (K⁺ 5.5–5.9 mmol/L) than current TA599. Relevant for managing patients on RAASi for CKD and heart failure. Final publication expected 29 April. Read the draft guidance.

QOF 2026/27 is live — four-pillar heart failure arrives. The new HF009 indicator (12 points) requires ACEi/ARB and beta-blocker and MRA and SGLT2 inhibitor in HFrEF. Obesity pharmacotherapy enters QOF for the first time via OB004/OB005 (18 points). Statin indicator points in diabetes have doubled (DM034) and quadrupled (DM035). Worth auditing your HFrEF list and statin cohorts this month.

PCN DES 2026/27 — opt-out deadline 30 April. The new DES raises the ARRS GP reimbursement ceiling to £118,759, retires the Advice & Guidance Enhanced Service, makes same-day urgent access mandatory, and adds RSV vaccination for care home residents and 80+ patients as a PCN obligation.

MHRA approves olezarsen (Tryngolza) for familial chylomicronemia syndrome. The APOC-III antisense oligonucleotide is approved for adults with FCS as adjunct to diet. Ultra-rare condition, specialist-initiated prescribing, but relevant for any practice holding a shared-care patient. Read the MHRA announcement.