Ambient voice technology in health care: what’s the evidence so far?

Ambient voice technology (AVT), sometimes called “AI scribes”, is rolling out across health care settings at pace. In England, as elsewhere, tools that listen to clinical consultations and automatically generate notes are being adopted in primary care, outpatient clinics, emergency departments and beyond. The promise is compelling: less time spent typing, more time for patients, and some relief from documentation pressures.

But while adoption is accelerating, robust evidence on what AVT actually delivers remains surprisingly limited.

This blog draws on Phase 1 findings from a national NIHR-funded evaluation of AVT in the NHS, led by the NIHR Rapid Service Evaluation Team. Phase 1 focused on reviewing the evidence, mapping the AVT market, and developing a logic model to understand how benefits are expected to be realised. You can explore the key findings and framework in our Phase 1 slide deck.

AVT can save documentation time

Across the published literature to date, one finding stands out: AVT can reduce the time that clinicians spend on documentation. Studies consistently show reductions in clinical time spent writing notes, and in some cases reductions in clinicians working out of hours on documentation, which echoes findings from other recent work where GPs reported that AI helps them reduce overtime work and the burden of admin. What we know much less about is what happens next.

Does time saved on documentation translate into more patient-facing care? Better continuity? Improved wellbeing? Increased capacity? Or does it simply get absorbed into an already overloaded system, smoothing the day but not fundamentally changing it?

Most existing evaluations stop at the point of documenting time savings. They rarely assess whether these savings lead to meaningful outcomes, such as patient experience, safety, workforce retention, or system capacity. In other words, we have some understanding of the outputs (minutes saved), but very little understanding of the outcomes (what those minutes enable).

This distinction matters. In health care, time is not automatically convertible into money, productivity, or better care. A clinician who saves two minutes per consultation may feel less pressured or less exhausted, but that does not automatically create an extra appointment slot or a cash-releasing saving. Without deliberate planning, time saved often becomes breathing space rather than measurable throughput, which may still be valuable, but in a different way.

A fast-moving and diverse market

The AVT market itself is evolving rapidly. Products are being adapted, reconfigured, and deployed across very different clinical contexts. Some tools focus narrowly on generating notes; others support letters, referrals, summaries or patient-facing outputs. Some are tightly integrated into electronic health records; others operate more loosely alongside them. New suppliers continue to enter the market, while established products are expanding their scope.

AVT is no longer confined to one type of consultation or one part of the system. It is being used or piloted across primary care, outpatient specialties, emergency care and community settings. That diversity makes evaluation harder, but also more important. Because tools and context vary so much, evidence drawn from one product or setting cannot be assumed to generalise to others.

The problem is not just lack of evidence but also inconsistent measurement

To date, how AVT has been evaluated varies greatly. Most studies rely on similar high-level indicators, such as “time in notes” or “total time spent in the electronic health record”, but define and measure them differently. Measures of wellbeing, job satisfaction, patient experience or safety are far less common, often bespoke, and rarely standardised. System-level impacts and costs are assessed in only a handful of studies.

This heterogeneity makes it difficult to compare findings across sites, products or settings. It also limits learning over time. Even where studies report positive impacts, it is often unclear whether differences reflect the technology, the context, the implementation approach, or simply how outcomes were measured.

Why a logic model matters

One way to address this problem is through a clear logic model. A logic model forces clarity about how a tool is expected to work: what goes in (the technology, training, workflow changes), what happens (how staff actually use it), what comes out (for example, reduced documentation time), and, crucially, what outcomes those outputs are meant to support.

For AVT, this clarity is essential. Reduced documentation time is not an outcome in itself; it is an intermediate step. The outcomes might be improved patient experience, reduced burnout, better quality records, improved safety, increased capacity, or better staff retention. But those outcomes do not occur automatically. They depend on how time is repurposed, how workflows change, and how the wider system responds.

Using a logic model enables a clearer articulation of AVT’s benefits, ensures that different forms of value (including wellbeing) are recognised, and supports shared expectations across clinicians, managers, suppliers and policymakers. It also allows consideration of potential unintended consequences – for example, increased documentation time initially, as transcribed notes and actions are double checked for accuracy. 

Looking ahead 

The first phase of this evaluation focused on laying foundations: mapping the existing evidence, understanding the market, and developing a logic model and a more consistent approach to measurement. It does not answer all the questions about AVT’s impact, but it clarifies which questions matter, and where current evidence falls short.

The next stage of the evaluation will build on this by undertaking a multi-site, mixed-methods evaluation of AVT in real-world NHS settings. It will examine not just whether AVT saves time, but how it is implemented, how it reshapes work, how staff and patients experience it, and whether anticipated benefits are actually realised. It will also explore economic implications and the conditions under which AVT might contribute to sustainable value for the health system.

As AVT continues to roll out, the challenge is not simply to move faster, but to learn faster, and more coherently. Clear logic, consistent metrics, and attention to outcomes rather than just outputs will be essential if AVT is to deliver on its promise for staff, patients and the system as a whole.