Cutting through the 'buzz' of ambient voice technology

Dr Devin Gray, GP and Clinical Lead for Digital Transformation at Wandsworth GP Federation, recently delivered a webinar for Wandsworth practices designed to cut through the “buzz” surrounding AVT and offer practical, actionable insights for primary care clinicians.

The main elements of the session are captured in this case study.

Her session demystified AVT, clarified crucial NHS standards, and aimed to equip practitioners with the knowledge to confidently explore and implement this technology safely and compliantly. The core focus was on safety, compliance, recent communications from NHSE, and contextualizing these within the real world of general practice.

Understanding Ambient Voice Technology

Navigating the evolving landscape of digital tools in primary care presents both exciting opportunities and significant challenges. Among these, Ambient Voice Technology (AVT) is rapidly gaining traction.

At its core, Ambient Voice Technology (AVT), often referred to as an AI Scribe, is a sophisticated tool designed to listen to natural patient consultations and convert spoken words into structured, summarized text. It functions by using advanced AI and Machine Learning algorithms to capture dialogue, identify medical terms, extract relevant data, and then produce outputs such as consultation summaries, referral letters, patient communications, and task lists. Some tools offer direct integration to populate fields in EMIS with a single click. Future ambitions for these tools include end-to-end automation, such as generating blood forms or prescriptions directly from the consultation.

Where does AI scribe technology fit into general practice? It is important to emphasise that AVT, like most other AI tools currently available to use in a healthcare context, must be considered an assistive technology[1]. Its purpose is to support the documentation process, enhance efficiency, and free clinicians to focus more on the patient. It does not, and should not, replace clinical judgment, active listening, or the nuanced human interaction that defines primary care.

Beyond timesaving: the broader impact of AVT

While timesaving is often the most cited benefit of AVT, I would strongly encourage clinicians and practices not to over-focus on this single aspect[2]. There are other equally, if not more, important benefits. For clinicians who typically document notes at the end of their appointments or even clinics, AVT can indeed save significant time. I haven’t ever managed to multitask typing during the consultation, and as result I estimate I save about 3 minutes per consultation. If one were to assume a BMA-compliant day of 25 patient contacts, that’s over an hour of time back each day. However, even for those who type during consultations, AVT can profoundly impact the relational aspect of care, allowing for more patient-focused interaction[3].

AVT can significantly enhance the quality of notes[4]. Multitasking is not cognitively possible, so it is inevitable that we miss moments of a patient’s narrative when we glance at the screen to follow-though on a thought process (“have they had a FIT test recently?”) or consider the next line of enquiry. An AI scribe, however, consistently captures details that a clinician might miss. The technology also tends to use neutral, non-judgemental and clear language, free from typos and spelling errors.

While AI scribes are not perfect and can “hallucinate” or omit information, the ability to review transcripts and edit summaries ensures accuracy, with the clinician always holding ultimate responsibility for what is saved in the patient record. Efficiency is also boosted through features like generated task lists, reducing the risk of missed actions and—if you’re anything like me—providing a welcome “dopamine hit” upon completion.

Perhaps the most powerful albeit hardest to measure benefit is the reduction in mental load. The cognitive task of recalling and structuring a consultation after the patient leaves consumes significant pre-frontal cortex power, contributing to decision fatigue[5]. Editing an AI-generated summary is cognitively much easier than starting from scratch. In my experience, this can lead to greater emotional tolerance for complex or tangential consultations, knowing key information is being captured. Ultimately, gaining even a little time back can help us to find that emotional reserve for more challenging encounters, and may in the long-term reduce the risk of burnout[6].

Am I allowed to us it?

The simple answer is yes; we are allowed to use this technology in the NHS. NHS England supports the responsible and safe adoption of AI tools that improve patient care and staff well-being. However, this comes with a crucial caveat: any AVT solution must meet a specific set of non-negotiable NHS standards, particularly concerning clinical safety and robust information governance. The responsibility to check and meet these standards ultimately sits with the individual practice as the deploying organization. Many practices are used to the idea of at least sharing the responsibility with the ICB or even the Federation, but this only applies to tools commissioned by these larger organisations. If you’re a GP practice buying an AVT tool for your practice or your PCN, the responsibility sits squarely with you.

Recent communications from NHSE’s National Chief Clinical Information Officer have brought the issue of compliance into sharp focus. A “priority notification” highlighted concerns that many AVT suppliers being used in the NHS do not meet the required standards, with a firm warning to “not use AVT solutions that are not compliant”[7]. While this caused a flurry of concerns shared in various forums and some degree of panic, it’s worth clarifying that this was not a change in guidance, but rather an emphatic reiteration that existing standards must be followed[8]. Whilst the liability for what is saved in a patient’s record has always and will always remain with the clinician, using non-compliant tools puts individuals and practices in a more precarious position if something were to go wrong[9].

Should I use it?

The biggest clinical risk when using AI scribe technology is over-reliance on the tool. AI can and does make mistakes. A recent study analysing nearly 13,000 clinician-annotated sentences summarised by Tortus AI demonstrated a 1.47% hallucination rate (the tool documented something that wasn’t said) and a 3.45% omission rate[10]. These figures are well below previously reported human note-taking rates. The risk to patients or indeed the clinician is very low if the output is carefully checked and edited to ensure it reflects the consultation. This should be emphasised in training when a practice is considering the use of this technology, and ongoing learning should be sought through audit and feedback in addition to existing incident management processes.

AVT as a medical device

A point of confusion has been the classification of AVT as a medical device with the Medicines and Healthcare products Regulatory Agency (MHRA), which was clarified to some extent by the priority notification. It is now clear that as AI scribes summarise text (rather than just transcribe or dictate) they are required to have Class I medical device status as a minimum, which is a self-assessment process. The CCIO was unequivocal in their position that we should only be using or even engaging with suppliers that have this in place. You can search the Public Access Registration Database (PARD) yourself, but it’s not always straightforward to find what you’re looking for, especially if the product is a result of a collaboration of companies[11]. Many of the larger AVT suppliers on the market are already working towards Class IIa, a more costly and lengthy process that will support future functionality aspirations.

Key compliance standards

To simplify what is required, practices considering the use of AVT must understand two main categories of standards:

Information Governance: This focuses on data security. Suppliers must meet or exceed the NHS Data Security and Protection Toolkit (DSPT) standards, have clear Data Processing Agreements (DPAs), and practices will need to complete their own Data Protection Impact Assessments (DPIAs).
Clinical Safety Standards: These relate to patientsafety. Two safety standards are established in law through the Health and Social Care Act 2012. These clearly state key roles and responsibilities, activities and documents required to undertake effective clinical risk management[12]. For any digital tool or IT solution that affects patient care in any way:
- Suppliers must comply with a safety standard called DCB0129[13]
- GP practices as the deploying organisation, must meet DCB0160[14].

The Digital Technology Assessment Criteria (DTAC) is a helpful, standardized summary of a supplier’s compliance, for both IG and clinical safety[15].

To meet the DCB0160 standard, the practice should appoint a lead (which both NHSE and CQC have stated must be a trained Clinical Safety Officer (CSO)^1,[16]) and conduct a local clinical risk assessment for the specific AVT solution. There are four specific documents that make up the DCB0160: a clinical risk management file, a clinical safety case report (where the main safety case for the tool is outlined and argued), a hazard log and an incident management file.

Choosing a supplier and implementation steps

Given the context on compliance, I recommend a “standards first, features second” approach when choosing a supplier. Non-negotiable requirements include NHS DSPT compliance, DCB0129 compliance, and MHRA Class I medical device registration. Beyond these, practices should evaluate accuracy, user experience, cost-effectiveness, scalability, support, and whether the supplier can support the practice’s own processes in meeting the DCB0160 standard. Consider whether the tool integrates with your electronic health record, and how it does so. It is still early days with integration in my view, with issues related to coded data, formatting of the consultation type and the risk that a completely seamless integration may encourage over-reliance on the tool.

The price structure of different AVT solutions varies from licensing to per-registered-patient models. Some offer individuals to sign-up and pay, and others only work with whole organisations or even only through existing digital suppliers as integrated partners (such as Accurx and X-On). On a per-registered-patient basis, the lowest price for a compliant tool that I’ve seen is 24p with Lexacom (but not with all their existing features), the highest is Accurx Scribe at 40p and most of the others are coming in at around 32p per patient.

For practices looking to try out AVT, here is a 10-step plan to implementation: Appoint a lead and get them CSO-trained if not already[17] Choose a supplier with due diligence as above, including ensuring Class I medical device registrationContact your DPO for the latest approved DPIA, and read, complete, and sign itAsk the supplier for their DTAC Write or adapt template DCB0160 documents Write or adapt template AI scribe practice policyUpdate your privacy noticeConsider a pilot phase (many offer free versions and trials)Support your team to understand the risks (mainly, over-reliance) and benefitsEngage and inform patients through various communication channels including waiting room screens, posters, websites, social media, phone messages and batch-email.

Patient engagement and monitoring

Effective patient engagement is key for successful and ethical AVT deployment. Patients should be informed before the consultation that AVT is in use, with clear explanations of its purpose such as to allow the clinician to focus more on them. A clear and simple opt-out option must be provided, along with an explanation of the implications of opting out i.e. the clinician can revert to manual notetaking. Practices should be prepared to answer common patient questions regarding data security, who hears the recordings, and whether the technology will replace their doctor.

Finally, ongoing monitoring is essential. Practices should define success metrics, regularly review clinical safety risks and incidents, and assess the impact on efficiency, documentation quality, and the overall consultation experience. By carefully navigating these considerations, primary care can safely and effectively harness the transformative potential of ambient voice technology.

The future of clinical safety in primary care

I recently heard the phrase “Health tech companies are the new pharmaceuticals”. NHSE, I believe, are responding to an increasingly Wild West scenario, where numerous new AI companies are emerging and targeting individual clinicians directly, often bypassing traditional procurement processes. This means GPs may be unaware of their legal responsibility to verify supplier compliance and conduct their own formal risk assessments¹. NHSE has maintained that it is up to the deploying organization to ensure a tool meets standards, led by a CSO. A challenging position for busy primary care practices without the training, resources or support to undertake formal clinical risk management processes.

I am increasingly convinced that we need to build the capability to assess and manage clinical risk management processes for digital tools such as AVT in Primary Care. There is no value in turning to our equally under-resourced ICBs who do not have a function in this where they are not the commissioners. CSOs across London have been working together with Primary Care Transformation Teams within ICBs to draft template compliance documents for products already being used widely across the Capital. However, clinical risk management is an ongoing process that iterates when mistakes and errors happen, and need updating when the supplier develops new functionality within the product. Whilst these templates can support practices get off the starting-blocks with meeting the DCB0160 standard, GPs need to fully understand and own the risks so that they can meaningfully manage them and keep patient harm to a minimum.

There is hope for future assurance and procurement frameworks to simplify this process, as outlined in the recently published 10-year plan for the NHS[18]. However, we want to harness the transformative potential of AI tools such as AVT now. In Wandsworth, my plan is to develop a community of practice of trained CSOs across the borough to collaborate on clinical safety compliance for current and new digital tools. I imagine AVT will be at the top of the list.

[1] https://www.england.nhs.uk/long-read/guidance-on-the-use-of-ai-enabled-ambient-scribing-products-in-health-and-care-settings/

[2] https://catalyst.nejm.org/doi/full/10.1056/CAT.25.0040

[3] https://www.thieme-connect.com/products/ejournals/abstract/10.1055/a-2337-4739

[4] https://link.springer.com/article/10.1007/s11606-024-08924-2

[5] https://pmc.ncbi.nlm.nih.gov/articles/PMC7555063/

[6] https://www.thieme-connect.com/products/ejournals/abstract/10.1055/a-2461-4576

[7] https://www.gponline.com/gps-advised-stop-using-ai-scribes-without-key-assurances/article/1922151

[8] https://www.pulsetoday.co.uk/news/technology/gp-practices-told-to-appoint-trained-officer-before-using-ai-transcription-tools/

[9] https://bit.ly/MPSCasebookAI

[10] https://www.nature.com/articles/s41746-025-01670-7

[11] https://pard.mhra.gov.uk/

[12] https://www.england.nhs.uk/long-read/digital-clinical-safety-assurance/