Blog | Reflections on emerging evidence for AVT

Recent weeks have given me the opportunity to pause and reflect on two important contributions to the evidence base for ambient voice technology (AVT): the BMJ analysis on unintended consequences of ambient scribes and the phase 1 findings from the NIHR RSET rapid evaluation. Taken together, these publications offer a timely and balanced perspective on both the promise and the limitations of AI-enabled clinical documentation. I welcome this work; it provides an essential foundation for responsible, human-centred innovation.

Ambient scribes: benefits, risks, and unresolved questions

The BMJ analysis by Eccles et al. examines the rapid global growth of ambient scribes and the clinical pressures they are intended to address — particularly in primary care, where administrative load, burnout and cognitive strain are well documented. The authors acknowledge potential benefits, including reduced cognitive burden, improved real-time documentation and enabling clinicians to focus more on patients.

However, the paper also rightly emphasises several critical concerns:

Inaccuracies and hallucinations within generated summaries
Loss of tacit and contextual nuance, which clinicians use to inform decision-making
Automation bias, especially when outputs appear authoritative
Potential inequities related to accents, dialects or language diversity
Medico‑legal uncertainties, including responsibility for AI generated errors

For many clinicians, these points reflect lived experience: AI documentation tools can be enormously helpful, but only when accuracy, transparency and clear patient consent remain non-negotiable.

The value of the BMJ analysis lies in making these constraints explicit. They should be seen not as reasons to avoid AVT altogether, but as conditions that must be met if such tools are to be deployed safely in everyday practice.

The RSET phase 1 mapping: a much needed UK perspective

In parallel, the RSET evaluation presents the most comprehensive mapping of AVT deployment in the UK to date. By reviewing 22 studies, creating a structured AVT taxonomy and charting 32 commercial products (14 already used in the NHS), the report reveals how fragmented, short term and methodologically inconsistent the current evidence landscape remains.

Their early findings highlight meaningful potential — reduced documentation burden, improved staff wellbeing, and renewed focus on patient interaction — but also underscore familiar challenges:

Persistent accuracy issues
Workflow friction and integration barriers
Inconsistent evaluation frameworks across settings

Crucially, the RSET logic model reinforces a central principle: AVTs deliver sustainable value only when they integrate smoothly into existing workflows, support clinician verification and enhance — rather than displace — the clinician–patient relationship.

A path forward: responsible, equitable, real world innovation

These publications make clear that ambient voice technologies must evolve with safety, equity and real‑world performance as the guiding metrics. I see enormous potential in tools that help clinicians reconnect with patients, reduce burnout and enhance documentation quality. Yet meaningful adoption demands rigor: transparent model behaviour, robust evaluation, inclusive design and continuous human oversight.

These insights strengthen our commitment to building AVT solutions that are accurate, explainable, aligned with UK regulatory frameworks and fundamentally designed around the needs of clinicians and patients.

Looking ahead: strengthening post deployment governance

As the NHS moves into the next phase of AVT implementation, we're keen to contribute to the development of stronger post-deployment surveillance. This must include transparent feedback loops that drive improvements not only in AVT performance but also in how these tools are used. Our goal is to amplify the positive impact of AVTs while ensuring that robust, transparent and compliant governance frameworks minimise unintended consequences.

About the author

Dr Lucy Mackillop

Chief medical officer - Data and Research

Dr Lucy Mackillop is Chief Medical Officer for Data and Research and Caldicott Guardian at Optum, as well as a practising Consultant Obstetric Physician and honorary senior clinical lecturer at the University of Oxford.