The use of ambient voice technology (AVT) for clinician–patient consultations in healthcare practice: A rapid systematic scoping review Authors: Rachel Lawrence, Kevin Herbert, Theo Georghiou, Raj Mehta, Pei Li Ng, Lucina Rolewicz, Stephen Morris, Jenny Shand Published: 2026 ABSTRACT ObjectiveUsing ambient voice technology (AVT) to transcribe and summarise clinician–patient consultations in real time is increasing in healthcare practice, with the aim of reducing clinical documentation burden. However, the evidence base remains underdeveloped. This review synthesised current evidence on the use and impact of AVT for documenting clinician–patient consultations in healthcare. We also examined how impact has been evaluated and measured with the aim of informing future research and assessment in real-world practice.MethodsWe conducted a rapid systematic scoping review (PROSPERO: CRD420251155579). Four databases and grey literature sources were searched in September 2025. Of 598 articles screened, 21 met the inclusion criteria. These studies evaluated user experiences with AVT (n=13), quantified impacts using objective numerical data (n=19) and perceived impacts based on self-reported experiences (n=17).ResultsEvidence on the real-world impact of AVT remains limited. While some studies report reduced documentation burden and improvements in staff well-being and perceived care quality, these findings largely stem from short-term observational designs and self-reported measures. Evidence relating to wider, system-level impacts such as pathway productivity, patient safety and cost effectiveness is sparse. Approaches to evaluating AVT were heterogeneous and inconsistent, particularly when reporting on system-level impacts beyond the clinician–patient consultation. Different factors (e.g., workflow integration, output quality), the consultation context and clinical setting influenced experiences of using AVT.ConclusionsCurrent evidence does not yet support widespread adoption of AVT in clinical practice. Although early findings suggest potentially important benefits, available studies are heterogeneous, predominantly short-term and methodologically limited. It remains unclear whether the improvements reported to date translate into sustained gains in clinical capacity, care quality or system efficiency. Longer-term evaluations using standardised outcome measures complemented by qualitative methods, are needed to determine the impact and value of AVT in healthcare.PROSPERO registration numberCRD420251155579. WHAT IS ALREADY KNOWN ON THIS TOPIC Use of ambient voice technology (AVT) in healthcare for documenting patient consultations in real time is growing rapidly, but the evidence base remains underdeveloped. Existing reviews have not focused specifically on AVT use in routine clinical practice and on how impact has been assessed and evaluated. Use of ambient voice technology (AVT) in healthcare for documenting patient consultations in real time is growing rapidly, but the evidence base remains underdeveloped. Existing reviews have not focused specifically on AVT use in routine clinical practice and on how impact has been assessed and evaluated. WHAT THIS STUDY ADDS This review shows that evidence on the real-world impact of AVT remains sparse. Most studies are short-term, methodologically varied and report on the perceived benefits rather than robust, quantitative outcomes. Measures of impact are inconsistent, particularly for wider, system-level outcomes such as productivity, safety and cost. This review shows that evidence on the real-world impact of AVT remains sparse. Most studies are short-term, methodologically varied and report on the perceived benefits rather than robust, quantitative outcomes. Measures of impact are inconsistent, particularly for wider, system-level outcomes such as productivity, safety and cost. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Current evidence does not yet justify widespread adoption of AVT. While some studies do suggest reductions in documentation burden and improvements in staff well-being and perceived care quality, findings rely largely on short-term observational designs and self-reported measures. More rigorous, standardised and longer-term evaluations are needed to determine whether AVT can deliver meaningful and sustained improvements at the system level. Current evidence does not yet justify widespread adoption of AVT. While some studies do suggest reductions in documentation burden and improvements in staff well-being and perceived care quality, findings rely largely on short-term observational designs and self-reported measures. More rigorous, standardised and longer-term evaluations are needed to determine whether AVT can deliver meaningful and sustained improvements at the system level. Introduction The use of digital innovations to improve the efficiency of healthcare services is rapidly advancing to address increasing demands, workforce shortages and care complexity.1–3 Artificial intelligence (AI), described as computer systems which perform tasks usually associated with human intelligence,4 is one innovation which has been described as potentially transformative for improving healthcare services.2 3 5–9 The optimism surrounding AI is also reflected in health policy,6 where AI is often the centre of future plans for improving service provision, patient outcomes and reducing workforce burnout.10 Ambient voice technology (AVT) is an AI technology that records clinician–patient consultations and transcribes and summarises the content discussed in real-time to automatically populate clinical documentation (e.g., patient note summaries, discharge letters).3 11 Different terminology can be used to describe ‘AVT’ within clinical and academic literature, including AI-scribe, ambient scribes and/or digital scribes. In this review, AVT will be used as an umbrella term encompassing these differences, aligning with ongoing policy10 communities of practice12 and evaluation work.13 14 Currently, the use of AVT in clinical practice is increasing across different settings (e.g., primary care, outpatients, inpatients, emergency care and mental health).15 16 The rationale for increased use is to reduce clinical documentation burden for the healthcare workforce.2 3 11 For example, evidence suggests that clinicians spend a significant amount of time on documentation tasks, often outside of usual working hours.2 3 11 17 Consequently, documentation burden can impact the quality of documentation and clinicians’ well-being, work-life balance and ability to focus on patient care to improve care quality.2 3 11 17 Reducing such burden by AVT automating certain tasks (e.g., summarising consultations so that clinicians only need to review/edit) is therefore a timely and promising innovation.3 16 18 However, integrating AVT into clinical practice can also come with challenges/risks, such as errors in documentation (e.g., hallucinations), misinterpretation which can impact documentation quality and the complexity of integrating AVT into existing workflows.18 19 Evaluating the real-world impact of AVT remains an evolving area of research, with emphasis placed on the importance of building the evidence base and understanding long-term productivity gains.3 7 16 To facilitate this, it is essential to understand what is already known about the use and impact of AVT. Existing reviews have highlighted potential benefits but also emphasised substantial evidence gaps, particularly in relation to long-term effectiveness, safety and economic value.2 17 20 Building on previous reviews and to inform future implementation and evaluation, we aimed to clarify how the impact of AVT has been measured to date, and what evidence exists on its effects in routine clinical practice. This review addressed the following research questions: How has the impact (quantified and perceived) of AVT on documenting clinician–patient consultations been evaluated (i.e., how has it been measured and defined)? How has the impact (quantified and perceived) of AVT on documenting clinician–patient consultations been evaluated (i.e., how has it been measured and defined)? What evidence exists on the impact (quantified and perceived) of using AVT to document clinician–patient consultations? What evidence exists on the impact (quantified and perceived) of using AVT to document clinician–patient consultations? What evidence exists on user experience with AVT to document clinician–patient consultations? What evidence exists on user experience with AVT to document clinician–patient consultations? Methods Registration This review was registered on PROSPERO (CRD420251155579). Design We conducted a rapid systematic scoping review,21 following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA 2020) guidelines (online supplemental file 1).22 23 A rapid scoping approach was chosen24 because of the breadth of topics covered (methods of evaluation, experiences of use and impact of AVT) and the emerging nature of the research area.7 16 19 This review forms part of a broader, two-phase mixed methods evaluation that also included the development of an AVT taxonomy, a market map of AVT products and a logic model (a framework linking AVT inputs and activities to outputs, short-term outcomes and longer-term impacts).13 The AVT taxonomy was used to help guide what was considered AVT in this review (online supplemental appendix 1), to help navigate varying terminology and descriptions of these technologies. The key components, covering transcription, summarisation and other product functions were used to help guide the information extracted and synthesised about AVT. Search strategy and selection criteria Studies were eligible for inclusion if they examined AVT tools for documenting clinician–patient consultations—defined as AI-assisted voice technologies that record, transcribe and generate summary outputs of consultations for review. Studies focusing solely on traditional dictation tools without an AI-assisted summary were excluded. We also restricted inclusion to studies evaluating AVT in real-world clinical practice. Studies conducted retrospectively, theoretically or in simulated scenarios were not eligible. Detailed inclusion and exclusion criteria are presented in table 1. Table 1Review inclusion and exclusion criteriaInclusion criteriaExclusion criteria Population and setting: International studies focusing on the use of AVT within routine clinical healthcare practice (including primary, secondary, community or home care settings). All population groups.Intervention: AI-assisted AVT that records, transcribes and generates clinical documentation from clinician–patient consultations. Tools limited to dictation or speech-to-text without AI-assisted summarisation were excluded.Study focus: Empirical evaluation of the real-world use, implementation and/or impact of AVT for documenting clinician–patient consultations.Study design: Primary empirical research (quantitative, qualitative or mixed methods), including pilot and quality improvement studies, with aims, methods and findings.Publication characteristics: Published in English from 2015 onwards (due to rapid AI advancements and pace of research10 19 55), in peer-reviewed journals or relevant grey literature (e.g., NHS reports or case studies). Studies could be provided in any format (including preprints) but needed to report on empirical results (as above).International evidence published in English. Population and setting: Studies conducted in simulated, retrospective, experimental or hypothetical settings rather than routine clinical practice. Studies focused solely on technical development, model training, algorithm validation or performance benchmarking without evaluation of real-world clinical use.Intervention: Studies examining transcription or dictation tools without AI-assisted summarisation functionality. Comparative modelling, prediction or workflow simulation studies that did not evaluate implementation in practice.Study focus: Studies not focused on AVT for documenting clinician–patient consultations.Study design: Studies without empirical findings.Publication characteristics: Review articles, commentary/opinion papers, editorials, retracted articles, conference abstracts and protocols. Studies published before 2015 and not published in English. AI, artificial intelligence; AVT, ambient voice technology; NHS, National Health Service. Review inclusion and exclusion criteria Population and setting: International studies focusing on the use of AVT within routine clinical healthcare practice (including primary, secondary, community or home care settings). All population groups. Population and setting: International studies focusing on the use of AVT within routine clinical healthcare practice (including primary, secondary, community or home care settings). All population groups. Intervention: AI-assisted AVT that records, transcribes and generates clinical documentation from clinician–patient consultations. Tools limited to dictation or speech-to-text without AI-assisted summarisation were excluded. Intervention: AI-assisted AVT that records, transcribes and generates clinical documentation from clinician–patient consultations. Tools limited to dictation or speech-to-text without AI-assisted summarisation were excluded. Study focus: Empirical evaluation of the real-world use, implementation and/or impact of AVT for documenting clinician–patient consultations. Study focus: Empirical evaluation of the real-world use, implementation and/or impact of AVT for documenting clinician–patient consultations. Study design: Primary empirical research (quantitative, qualitative or mixed methods), including pilot and quality improvement studies, with aims, methods and findings. Study design: Primary empirical research (quantitative, qualitative or mixed methods), including pilot and quality improvement studies, with aims, methods and findings. Publication characteristics: Published in English from 2015 onwards (due to rapid AI advancements and pace of research10 19 55), in peer-reviewed journals or relevant grey literature (e.g., NHS reports or case studies). Studies could be provided in any format (including preprints) but needed to report on empirical results (as above). Publication characteristics: Published in English from 2015 onwards (due to rapid AI advancements and pace of research10 19 55), in peer-reviewed journals or relevant grey literature (e.g., NHS reports or case studies). Studies could be provided in any format (including preprints) but needed to report on empirical results (as above). International evidence published in English. International evidence published in English. Population and setting: Studies conducted in simulated, retrospective, experimental or hypothetical settings rather than routine clinical practice. Studies focused solely on technical development, model training, algorithm validation or performance benchmarking without evaluation of real-world clinical use. Population and setting: Studies conducted in simulated, retrospective, experimental or hypothetical settings rather than routine clinical practice. Studies focused solely on technical development, model training, algorithm validation or performance benchmarking without evaluation of real-world clinical use. Intervention: Studies examining transcription or dictation tools without AI-assisted summarisation functionality. Comparative modelling, prediction or workflow simulation studies that did not evaluate implementation in practice. Intervention: Studies examining transcription or dictation tools without AI-assisted summarisation functionality. Comparative modelling, prediction or workflow simulation studies that did not evaluate implementation in practice. Study focus: Studies not focused on AVT for documenting clinician–patient consultations. Study focus: Studies not focused on AVT for documenting clinician–patient consultations. Study design: Studies without empirical findings. Study design: Studies without empirical findings. Publication characteristics: Review articles, commentary/opinion papers, editorials, retracted articles, conference abstracts and protocols. Studies published before 2015 and not published in English. Publication characteristics: Review articles, commentary/opinion papers, editorials, retracted articles, conference abstracts and protocols. Studies published before 2015 and not published in English. AI, artificial intelligence; AVT, ambient voice technology; NHS, National Health Service. Four databases and the grey literature were searched on September 29th 2025: MEDLINE-Ovid (PubMed), Web of Science Core Collection, CINAHL Plus and EMBASE. Grey literature was identified using healthcare policy websites (e.g., National Health Service England, The Health Foundation, Incubator for AI and Digital Health), reference lists of existing reviews2 20 and grey literature platforms (e.g., Google Scholar and Google search). One researcher conducted the search and imported all records into EndNote25 and Rayyan26 to remove duplicates. We iteratively developed the search strategy using MEDLINE-Ovid, assisted by a University College London librarian, who provided support with identifying key terminology associated with AVT and mapping search terms to broader subject headings. Search terms were also informed by prior research and existing AI-focused reviews.1 2 Terms covered the following concepts: AI and machine learning, AI scribes, ambient voice technology and documentation (online supplemental appendix 2). Screening and study selection Studies were screened in three stages: (1) title; (2) abstract and/or study summaries; and (3) full texts.21 24 One researcher (RL) screened all titles and abstracts and 10% of excluded papers were independently reviewed by one of two researchers (KH/TG).24 Three researchers independently screened all full texts (RL/KH/TG), with regular discussions throughout the screening process.24 Any disagreements in relation to exclusions were resolved through discussions between researchers (RL/KH/TG) until consensus was reached. Data extraction and synthesis A structured data extraction tool was developed to capture key findings from all included studies. Extracted data included: study characteristics; type and functionality of AVT tools; implementation context and setting; study methods, metrics and design; findings on user experiences; and evidence of AVT impact on: (1) staff; (2) patients; and (3) providers/the healthcare system (i.e., system-level stakeholders and healthcare organisations). The tool was piloted on 10% of included studies (n=2) by two researchers (RL/KH) in line with rapid review guidance21 and no changes were made as a result of this process. Any disagreements during the development or refinement of the tool were resolved through team discussion until consensus was reached. All data relevant to the review questions was extracted, including illustrative quotes from qualitative studies. Given the heterogeneity of study designs, neither quantitative meta-analysis nor formal qualitative synthesis was feasible—an issue also noted in previous reviews of AVT.2 17 We therefore conducted a narrative synthesis to integrate and analyse findings across all studies.27 This approach allowed us to incorporate both the perceived impact (self-reported experiences from clinicians and/or patients) and quantified impact (using objective, numerical data such as electronic health record (EHR) data and/or operational metrics). We descriptively summarised study characteristics, the types of AVT tools used and reported user experiences. The key factors influencing user experiences were grouped into barriers and facilitators and frequencies were reported across studies. Findings related to AVT impact, and the measures used to assess impact, were synthesised thematically and grouped by outcomes for (1) staff; (2) patients; and (3) providers/the healthcare system. Where quantitative studies reported estimates of the scale of impact (e.g., percentage change compared with a comparator), these values were taken directly from the individual studies; they do not represent pooled meta-analytic effect sizes. All findings were discussed and agreed across the wider team. Critical appraisal We assessed the quality of included studies using the Mixed Methods Appraisal Tool (MMAT), a tool designed for the evaluation of qualitative, quantitative and mixed methods designs within the same review.28 The tool was applied by individual researchers to qualitative studies (RL), descriptive surveys (RL) and quantitative studies (KH), while mixed methods studies were assessed collaboratively (RL/KH). Also, to provide an overall summary of study quality, studies were rated on a five-level grading system (‘very low’, ‘low’, ‘moderate’, ‘good’, ‘high’), following the guidance for reporting the results of the MMAT (V.2018).29 Any study failing to meet the screening criteria questions was assessed as ‘very low’. Qualitative and quantitative studies were then assessed based on the lowest score (proportion of criteria questions receiving a ‘yes’ rating) of the relevant study type components.29 For mixed methods studies, quality ratings could not exceed that of the lowest score from its key components: qualitative, quantitative, or mixed methods integration.29 Patient and public involvement Patient and public involvement and engagement (PPIE) team member (RM) was actively involved throughout key study phases, supporting the development of the review protocol and approach, providing feedback on data extraction tools and review analysis and write-up. The rapid evaluation team has a wider programme of PPIE leadership (RM and PLN), overseeing PPIE involvement across projects, aligning with evidence highlighting the importance of PPIE input to ensure research is relevant (for patients, the public and services), meaningful and of high quality.30 Results Study selection and characteristics A total of 885 records were identified with 21 meeting the inclusion criteria (figure 1, online supplemental file 2). Table 2 summarises study characteristics. Most studies were quantitative (n=11) or mixed methods (n=8), with only two qualitative studies. The majority (n=17) were conducted in the USA, with only one from the UK. AVT was implemented across a range of health settings, with studies often conducted within an integrated health system or academic medical centre spanning multiple settings (n=9). Most studies evaluated AVT within short-term pilots (n=8) or quality improvement initiatives (n=6) typically lasting 0–6 months. Study population groups commonly included clinicians (n=13), clinicians and patients (n=7) or patients only (n=1). PRISMA flow diagram. AI, artificial intelligence; AVT, ambient voice technology; NHS, National Health Service; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analysis; THIS, The Healthcare Improvement Studies Institute. Table 2Summary of study characteristicsStudy characteristicsTotal (out of n=21 overall)*Study focus Experiences of using AVT13 Quantified impact of AVT19 Perceived impact of AVT17Outcome/s reported† – impact on staff Documentation burden19 Documentation quality12 Staff well-being15Outcome/s reported† – impact on patients Care quality14Outcome/s reported† – impact on provider/healthcare system Wider pathway productivity5 Patient safety2 Cost2Country USA17 UK1 Spain1 Canada1 Australia1Methodological approach – overall Quantitative11 Mixed methods8 Qualitative2Study design – quantitative Prospective study3 Pre–post observational study2 Randomised clinical trial1 Randomised controlled trial (stepped-wedge design)1 Single-group study1 Peer-matched controlled cohort study1 Cross-sectional survey1 Prospective observational study1Study design – mixed methods Pre–post study (including open-ended survey responses)3 Pre–post survey (with subset of interviews)2 Quasi-experimental design (including survey with open-ended element)1 Pre–post study with survey and interviews1 Cross-sectional survey1Study design – qualitative Semi-structured interviews2Study setting Across an integrated health system or academic medical centre (across multiple settings)9 Outpatient care in an integrated health system or academic medical centre4 Primary care3 Community dermatology1 Community palliative medicine1 Paediatric otorhinolaryngology (ENT) department1 Outpatient paediatric provider1 Family medicine across an integrated system1Implementation timeframe Pilot (0–6 months)8 Quality improvement study (0–6 months)6 Unclear4 Pilot (6 months to 1 year)2 Pilot (1 year)1Population group studied Clinicians13 Clinicians and patients7 Patients only1*All numbers are total reported/21 as the denominator.†Some studies reported on multiple different outcomes. Totals for each outcome type (e.g., impact on staff, impact on patients) may not always total 21.AVT, ambient voice technology. Summary of study characteristics *All numbers are total reported/21 as the denominator. †Some studies reported on multiple different outcomes. Totals for each outcome type (e.g., impact on staff, impact on patients) may not always total 21. AVT, ambient voice technology. Type of AVT tools All studies evaluated tools which used generative voice technology driven by AI. Most studies evaluated Dragon Ambient eXperience (DAX) Copilot.31–42 Other AVT systems studied included Lyrebird scribe,43 Abridge,34 ScribeAmerica (Telescribe and Speke),44 Google Gemini,45 Nabla,36 Tortus AI15 and Scribeberry.46 Two studies34 36 used more than one AVT tool. In four studies, the specific product was not named although its functionality was described.47–50 The level of detail provided on AVT functions and integration into clinical workflows varied considerably across studies. Functions of AVT tools Six studies described the use of note templates or structured formats.15 33 35 37 41 46 Three reported templates organised into four sections: (1) history of illness, (2) physical examination results, (3) assessment and (4) plan.33 37 41 Other studies used standardised templates15 35 46 often tailored using anonymised clinical notes15 or clinician prompts.46 Across studies the transcription-to-summary process was consistently described whereby AVT systems recorded clinician–patient consultations, produced transcribed text and generated AI-assisted summary outputs for clinician review.15 31 33 35 37 40 41 43–45 47–49 In the majority of studies, AVT was integrated into existing workflows to facilitate this process (i.e., embedded within a health systems EHR to populate patient notes).31–34 36–39 41–44 49 50 In other studies, AVT operated alongside existing systems through web-based applications, enabling AVT outputs to be integrated into EHRs after being reviewed/edited.15 35 45 47 48 In a small number of studies integration was not clearly described.40 AVT-generated summaries were made available to clinicians automatically after the consultation.15 31 33 37 44 45 48 49 Clinicians were required to review and edit the summarised notes to ensure they were accurate and clinically valid, before the documentation was finalised.15 31 33 35 41 43–49 Patient consent Patient consent processes were not consistently reported. Ten studies documented consent procedures—either written or verbally—prior to AVT use.15 31 33 35 43 44 46–49 Other studies notified patients that consultations would be recorded and reviewed by clinicians but did not explicitly mention AVT,41 or incorporated AVT within broader routine consent statements covering digital innovation.39 RQ1: Evaluating impact (quantified and perceived) A range of measures were used to evaluate the quantitative impact of AVT (table 3), with notable inconsistencies in study design and methodological approaches. Documentation time during and outside of working hours (by period, documentation item or activity type) was the most common metric for evaluating the impact on documentation burden. These measures typically focused on EHR-derived indicators (e.g., time in notes, total EHR time, after-hours work) and were assessed using pre–post comparisons or user versus non-user comparisons. Validated instruments for assessing staff well-being were used infrequently and inconsistently (examples included National Aeronautics and Space Administration (NASA) Task Load Index, Mini-Z burnout instrument). Quantitative evaluation of AVT’s impact on patient care quality, documentation quality and outcomes for providers/the healthcare system was less common. Perceived impact was evaluated through interviews or surveys, with findings reported descriptively or through statistical comparison of self-reported measures. Table 3Summary of measures used to evaluate the impact of AVT (quantified and perceived)OutcomeMeasures of quantified impactDenominator descriptionsMeasures of perceived impact Staff Documentation burden Time spent by physician/clinician working—in whole or in part—on documenting patient care (e.g., clinical note data entry, letters, orders for medications, tests or procedures)Total time spent on documentation workload (working hours): Documentation time/time in notes/time in system (n=10).15 33 34 36–38 42 46 48 49 EHR/EMR time (n=4).32 35 37 47 Time spent on specific documentation tasks (e.g., clinical notes, orders, letters, in-basket) (n=3)43 48 49 Total time spent on documentation workload (out of hours):Pyjama time (n=4).33 42 48 49 Time outside of working/scheduled hours (n=8).35 37 38 42 43 47–49 Time on unscheduled days (n=2).48 49 Time saved on documentation workload (in and out of hours):Total time saved (n=4).37 44 47 49 Varied between reports:Defined timeframes (e.g., per day, week, year) (n=8).32 33 37 42–44 46 47 Documentation item (e.g., per note) (n=4).15 34 36 37 Unit activity (e.g., encounter, appointment, patient) (n=8).15 33 35 38 42 47–49 Interviews (n=5).15 31 40 43 44 Surveys (with descriptive quantitative and/or qualitative data) (n=8).15 32 33 41 43 44 47 50 Surveys (with statistical comparisons) (n=5).15 33 42 43 50 Documentation quality Degree to which clinical documentation reflects the patient’s condition, care and outcomes in terms of, e.g., accuracy, precision, completeness and consistencyBERTscore (semantic relevance and response accuracy) (n=1)45 PDQI-9 scores (n=2)45 48 Overall note quality.By domain: thoroughness, organised, internally consistent, comprehensiveness, timeliness, clarity, conciseness relevance, usefulness for communication. N/A Interviews (n=3).15 31 40 Surveys (with descriptive quantitative and/or qualitative data) (n=8).15 32 33 41 45 47 48 50 Surveys (with statistical comparisons) (n=5).15 36 41 45 50 Clinician Likert ratings (directly after editing output) (n=1).46 Staff well-being Clinician-reported burnout, high task load levels, low engagement or professional fulfilmentScore vs comparator:NASA Task Load Index (n=2).15 41 Mini-Z Burnout Instrument (n=1).36 PFI (n=3).36 41 50 OLBI (n=1).38 Press Ganey Workforce engagement survey (n=1).35 N/A Interviews (n=5).15 31 40 43 44 Surveys (with descriptive quantitative and/or qualitative data) (n=6).15 32 33 44 47 49 Surveys (with statistical comparisons) (n=5).15 33 42 47 50 Patients Care quality Clinician–patient relationship and experience of care provision (e.g., being more patient focused, less typing) PDRQ-9, Wilcoxon rank-sum test (n=1).39 N/A Interviews (clinicians) (n=5).15 31 40 43 44 Interviews (patients) (n=1).43 Surveys (with descriptive quantitative and/or qualitative data) (n=6).15 33 47–50 Surveys (with statistical comparisons) (n=6).15 33 36 39 42 47 Surveys including patient perspectives (n=7).15 32 39 43 47–49 Providers/healthcare system Wider pathway productivity Impact of AVT on the wider pathway beyond the clinician-patient consultationSystem efficiency:Care process completion rates (n=4)15 33 35 43 and turnaround times (n=1).15 System capacity:Number of patients seen/assessed/treated within a given time period (n=1).15 Patient flow:Staff time on scene (n=1).15 Time between points on the care pathway (n=1).15 N/A Interviews (n=2).15 44 Surveys (with statistical comparisons) (n=1).33 Patient safety Safety of care, e.g., adverse events, any implications of AVT use on safety outcomes Safety events reported during the intervention period (n=2).35 36 Evidence of patient safety-related shortcomings in AVT outputs (e.g., overcoding, bias clinically significant inaccuracies) (n=2).35 36 N/AN/A Cost Costs or cost savings of AVT use Cost savings: Estimations to be achieved (n=2).15 32 Estimation:Compared with in-person scribe (supporting administrative staff) (n=1).32 Estimation (clinician, trust and national level) (n=1)15:Via:Reduced documentation time.Additional clinical capacity.Per:Documentation task.Shift.Week.Year. N/ANote: Some publications reported more than one outcome measure and/or used different denominators. Some studies also measured quantified and perceived impacts.AVT, ambient voice technology; BERTscore, Bidirectional Encoder Representations from Transformers; EHR, electronic health records; EMR, electronic medical records; N/A, not applicable; NASA, National Aeronautics and Space Administration; OLBI, Oldenberg Burnout Inventory survey; PDQI, Physician Documentation Quality Instrument; PDRQ, Patient–Doctor Relationship Questionnaire; PFI, Professional Fulfilment Index. Summary of measures used to evaluate the impact of AVT (quantified and perceived) Documentation time/time in notes/time in system (n=10).15 33 34 36–38 42 46 48 49 Documentation time/time in notes/time in system (n=10).15 33 34 36–38 42 46 48 49 EHR/EMR time (n=4).32 35 37 47 EHR/EMR time (n=4).32 35 37 47 Pyjama time (n=4).33 42 48 49 Pyjama time (n=4).33 42 48 49 Time outside of working/scheduled hours (n=8).35 37 38 42 43 47–49 Time outside of working/scheduled hours (n=8).35 37 38 42 43 47–49 Time on unscheduled days (n=2).48 49 Time on unscheduled days (n=2).48 49 Total time saved (n=4).37 44 47 49 Total time saved (n=4).37 44 47 49 Defined timeframes (e.g., per day, week, year) (n=8).32 33 37 42–44 46 47 Defined timeframes (e.g., per day, week, year) (n=8).32 33 37 42–44 46 47 Documentation item (e.g., per note) (n=4).15 34 36 37 Documentation item (e.g., per note) (n=4).15 34 36 37 Unit activity (e.g., encounter, appointment, patient) (n=8).15 33 35 38 42 47–49 Unit activity (e.g., encounter, appointment, patient) (n=8).15 33 35 38 42 47–49 Interviews (n=5).15 31 40 43 44 Interviews (n=5).15 31 40 43 44 Surveys (with descriptive quantitative and/or qualitative data) (n=8).15 32 33 41 43 44 47 50 Surveys (with descriptive quantitative and/or qualitative data) (n=8).15 32 33 41 43 44 47 50 Surveys (with statistical comparisons) (n=5).15 33 42 43 50 Surveys (with statistical comparisons) (n=5).15 33 42 43 50 Overall note quality. Overall note quality. By domain: thoroughness, organised, internally consistent, comprehensiveness, timeliness, clarity, conciseness relevance, usefulness for communication. By domain: thoroughness, organised, internally consistent, comprehensiveness, timeliness, clarity, conciseness relevance, usefulness for communication. Interviews (n=3).15 31 40 Interviews (n=3).15 31 40 Surveys (with descriptive quantitative and/or qualitative data) (n=8).15 32 33 41 45 47 48 50 Surveys (with descriptive quantitative and/or qualitative data) (n=8).15 32 33 41 45 47 48 50 Surveys (with statistical comparisons) (n=5).15 36 41 45 50 Surveys (with statistical comparisons) (n=5).15 36 41 45 50 Clinician Likert ratings (directly after editing output) (n=1).46 Clinician Likert ratings (directly after editing output) (n=1).46 NASA Task Load Index (n=2).15 41 NASA Task Load Index (n=2).15 41 Mini-Z Burnout Instrument (n=1).36 Mini-Z Burnout Instrument (n=1).36 PFI (n=3).36 41 50 PFI (n=3).36 41 50 OLBI (n=1).38 OLBI (n=1).38 Press Ganey Workforce engagement survey (n=1).35 Press Ganey Workforce engagement survey (n=1).35 Interviews (n=5).15 31 40 43 44 Interviews (n=5).15 31 40 43 44 Surveys (with descriptive quantitative and/or qualitative data) (n=6).15 32 33 44 47 49 Surveys (with descriptive quantitative and/or qualitative data) (n=6).15 32 33 44 47 49 Surveys (with statistical comparisons) (n=5).15 33 42 47 50 Surveys (with statistical comparisons) (n=5).15 33 42 47 50 PDRQ-9, Wilcoxon rank-sum test (n=1).39 PDRQ-9, Wilcoxon rank-sum test (n=1).39 Interviews (clinicians) (n=5).15 31 40 43 44 Interviews (clinicians) (n=5).15 31 40 43 44 Interviews (patients) (n=1).43 Interviews (patients) (n=1).43 Surveys (with descriptive quantitative and/or qualitative data) (n=6).15 33 47–50 Surveys (with descriptive quantitative and/or qualitative data) (n=6).15 33 47–50 Surveys (with statistical comparisons) (n=6).15 33 36 39 42 47 Surveys (with statistical comparisons) (n=6).15 33 36 39 42 47 Surveys including patient perspectives (n=7).15 32 39 43 47–49 Surveys including patient perspectives (n=7).15 32 39 43 47–49 Care process completion rates (n=4)15 33 35 43 and turnaround times (n=1).15 Care process completion rates (n=4)15 33 35 43 and turnaround times (n=1).15 Number of patients seen/assessed/treated within a given time period (n=1).15 Number of patients seen/assessed/treated within a given time period (n=1).15 Staff time on scene (n=1).15 Staff time on scene (n=1).15 Time between points on the care pathway (n=1).15 Time between points on the care pathway (n=1).15 Interviews (n=2).15 44 Interviews (n=2).15 44 Surveys (with statistical comparisons) (n=1).33 Surveys (with statistical comparisons) (n=1).33 Safety events reported during the intervention period (n=2).35 36 Safety events reported during the intervention period (n=2).35 36 Evidence of patient safety-related shortcomings in AVT outputs (e.g., overcoding, bias clinically significant inaccuracies) (n=2).35 36 Evidence of patient safety-related shortcomings in AVT outputs (e.g., overcoding, bias clinically significant inaccuracies) (n=2).35 36 Estimations to be achieved (n=2).15 32 Estimations to be achieved (n=2).15 32 Compared with in-person scribe (supporting administrative staff) (n=1).32 Compared with in-person scribe (supporting administrative staff) (n=1).32 Via:Reduced documentation time.Additional clinical capacity. Via: Reduced documentation time. Reduced documentation time. Additional clinical capacity. Additional clinical capacity. Per:Documentation task.Shift.Week.Year. Per: Documentation task. Documentation task. Shift. Shift. Week. Week. Year. Year. Note: Some publications reported more than one outcome measure and/or used different denominators. Some studies also measured quantified and perceived impacts. AVT, ambient voice technology; BERTscore, Bidirectional Encoder Representations from Transformers; EHR, electronic health records; EMR, electronic medical records; N/A, not applicable; NASA, National Aeronautics and Space Administration; OLBI, Oldenberg Burnout Inventory survey; PDQI, Physician Documentation Quality Instrument; PDRQ, Patient–Doctor Relationship Questionnaire; PFI, Professional Fulfilment Index. RQ2: Evidence of impact Studies assessed the impact of AVT across outcomes for staff, patients and providers/the healthcare system, reporting a wide range of effects across these groups (table 4). Table 4Summary of impact outcomes reportedOutcomes reportedNumber of studies; total(n=21)Number of studies;quantitative impactNumber of studies;perceived impactStaff Documentation burden191511 Documentation quality12212 Staff well-being15811Patients Care quality14214 Providers/the healthcare system Wider pathway productivity543 Patient safety220 Cost220Note(s): Some studies reported both quantified and perceived impacts, so will not always add up to the total number. Summary of impact outcomes reported Note(s): Some studies reported both quantified and perceived impacts, so will not always add up to the total number. Staff Documentation burden The majority of studies quantifying documentation burden (n=12 out of 15) reported reductions with AVT use, with the magnitude of impact often dependent on the clinical context and the extent of workflow integration. Nominal reductions in documentation burden were observed to range between 1.7% and 9.5% for outpatient clinicians and between 11.8% and 33.0% for primary care providers. The remaining studies reported either a mixture of outcomes indicating reduced burden or inconclusive results,48 outcomes indicating reduced burden or inconclusive results, plus a single outcome indicating increased burden,49 or one outcome each indicating reduced burden or increased burden.35 Beyond noting the overall consistency of a generally positive impact from AVT implementation, the ability to directly compare study findings and draw causal inferences as to the basis for any differences is precluded by several factors. Across the relevant publications, there is a lack of consistency in study design (conducted predominantly over the short-term), often performed on single AVT products based in differing healthcare settings and using varied outcome denominators (e.g., defined timeframes, documentation item, unit activity) (table 3). Although model specification varied between studies, regression analyses indicated associations between AVT use and reduced documentation time.36 42 48 49 Reductions in out of hours documentation time were reported in most studies (n=14)—ranging from 12% to 30% where comparative estimates were reported.15 33 34 36–38 41–44 46–49 However, one study noted a 5% increase in the percentage of documentation time spent after hours.35 Several studies (n=11) reported perceived improvements in documentation efficiency when using AVT (e.g, less time spent on documentation/administrative notes, having enough time available for documentation).15 31–33 40–44 47 50 However, some (n=5) noted that substantial proofreading and editing requirements could offset documentation efficiency gains15 31 33 40 50; consequences which were not always factored into the overall assessment of documentation burden. Documentation quality Two studies reported quantitative outcomes for documentation quality, measuring Physician Documentation Quality Instrument (PDQI-9) for AVT use.45 48 These studies reported PDQI-9 domain scores over four (maximum score of 5), indicative of high note quality.45 48 Perceptions of documentation quality were mixed. Several studies (n=5) reported satisfaction with documentation quality,15 33 40 41 45 but there was limited41 46 or no evidence15 of positive changes in perceived note quality following AVT implementation (n=2 and n=1, respectively). A number of challenges were also identified in relation to documentation accuracy (e.g., AVT including incorrect or irrelevant details (n=8),15 31 33 36 40 45 48 50 hallucinating events (n=5)15 31 40 45 48 or oversimplifying details (n=3)15 36 47) and organisation/formatting issues (e.g., structuring notes incorrectly, such as not in chronological order, not in the right sections, misalignment with note template (n=8)15 31 33 36 45 47 48 50). Staff well-being Quantitative studies reported improvements in staff well-being (n=1)35 and reductions in burnout (n=7)15 36 38 41 42 47 50 following AVT implementation. Although typically based on short-term observational study designs, the impacts were statistically significant (p<0.05 or lower). Staff also reported perceived improvements in well-being (n=10),15 31–33 40 42–44 47 50 work-life balance (e.g., spending less personal time on documentation (n=3))31 40 47 and job satisfaction (n=2),47 49 noting that AVT alleviated the cognitive burden (n=4),15 31 33 43 stress (n=1)32 and burnout (n=6)15 33 40 42 44 47 associated with documentation. However, some studies (n=2) emphasised that AVT did not remove all aspects of cognitive burden associated with documentation.31 33 Patients Care quality Although causal attribution could not be established, in one study a 23.6% relative increase in the percentage of direct care (speaking to or examining the patient without multitasking) during consultations was reported after the introduction of AVT.15 Another study found no difference in a rank sum test of Patient–Doctor Relationship Questionnaire (PDRQ-9) responses between baseline and post-implementation periods.39 While objective evidence was limited, studies reported perceived improvements in consultation quality for both clinicians (n=13)15 31–33 36 40 42–44 47–50 and patients (n=6),15 32 39 43 48 49 noting reduced screen time/typing to make consultations more personable and patient focused. However, three studies reported either no changes in patient perceptions regarding clinician engagement (n=1),49 feeling listened to/trusting their clinician with their care (n=1)47 or overall satisfaction with their consultation (n=1)15 following AVT use. Providers/healthcare system Wider pathway productivity (beyond consultation) Studies reporting on the quantified (n=4)15 33 35 43 and perceived impact (n=3)15 33 44 of AVT on system capacity or patient flow indicated improvements following AVT implementation (e.g., patients seen in a given timeframe, time to next step of patient care pathway and/or indirect efficiency improvements—such as patients feeling listened to and cared for or expedited administrative processes). However, one study found increases in self-reported patient volume33 and clinicians emphasised that to date, AVT has not been shown to directly impact measurable clinical outcomes for patients.15 Patient safety Only two studies reported patient safety-related outcomes.35 36 Detected safety event rates during the AVT implementation period were low (<0.01%, n=15,252),36 and no increase in overcoding or clinically significant inaccuracies was reported relative to baseline documentation.36 However, the safety monitoring approaches in these studies were variably described, follow-up periods were short and given the rarity of events (only one patient safety event being reported across the two studies), clinically significant harms may not have been detectable within the study timeframes.35 36 Cost Although no time frame was reported, one study32 indicated potential savings from AVT use relative to an in-person scribe (supporting administrative staff), while another projected savings at clinician, organisational and national levels, based on time savings and assumed increases in clinical capacity.15 RQ3: Use of AVT in practice Thirteen studies reported findings related to experiences of using AVT in practice,10 25 27 30 34 35 37–39 41–44 including usability, factors influencing use and the contexts in which clinicians thought AVT worked well. Seven studies reported that AVT was an easy and useful tool to use.15 33 36 40 41 44 48 AVT could facilitate efficient real-time documentation (n=3)15 33 41 while maintaining clinical detail (n=1)33 and reducing post-consultation administrative tasks (n=1).15 However, AVT outputs often required editing (n=5)15 31 33 40 45 and did not always align with, or improve, clinicians’ established documentation processes (n=3).15 31 36 The importance of clinicians’ maintaining their clinical style and thought process in AVT outputs was emphasised (n=2),15 40 to ensure clear outputs for patients (n=2)15 44 and maintain accountability (n=1).15 Patients were generally receptive to AVT use (n=6),15 36 39 40 43 47 49 though the importance of transparency regarding data use was noted (n=3).15 43 47 A number of factors influenced experiences of using AVT in routine practice (table 5). Table 5Influencing factorsFacilitatorsBarriers Integration into existing EHR workflows (n=4).15 44 48 49 Ease of use and editing (n=2).40 44 Data transparency (n=2).15 43 Previous positive experiences with AVT (n=2).44 49 Technology leaders and/or champions to support adoption and an openness to innovation (n=2).48 49 Well designed, personalised note templates to enable usability (n=1).15 Sufficient investment, training and practical support (n=1).15 Challenges with documentation quality (e.g., in relation to accuracy, detail, length and/or structure of AVT outputs) (n=7).15 31 36 40 45 47 48 Varied effectiveness across different settings, types of consultations (n=4)44 48–50 or scribes (n=1).44 Lack of integration with current workflow and/or existing documentation processes (n=3).15 48 49 Technical challenges with accessing and using the tools (n=4).15 40 44 48 Concerns about data privacy (n=2).15 47 Lack of familiarity with AVT/readiness for innovation (n=2).15 48 AVT tools not impacting documentation time (i.e., clinicians feeling they could type the notes more quickly than receiving and editing an AVT output) (n=2).31 49 Lack of personalised template designs which align reality with AVT output (n=2).15 47 Undefined approaches and limited consistency when evaluating the accuracy and safety of AVT in practice (n=1).48 Environmental factors (e.g., noise) and tool reliability (n=1).15 Unintended consequences AVT implementation could result in clinicians being expected to increase patient volume (n=2).15 31 AVT, ambient voice technology; EHR, electronic health record. Influencing factors Integration into existing EHR workflows (n=4).15 44 48 49 Integration into existing EHR workflows (n=4).15 44 48 49 Ease of use and editing (n=2).40 44 Ease of use and editing (n=2).40 44 Data transparency (n=2).15 43 Data transparency (n=2).15 43 Previous positive experiences with AVT (n=2).44 49 Previous positive experiences with AVT (n=2).44 49 Technology leaders and/or champions to support adoption and an openness to innovation (n=2).48 49 Technology leaders and/or champions to support adoption and an openness to innovation (n=2).48 49 Well designed, personalised note templates to enable usability (n=1).15 Well designed, personalised note templates to enable usability (n=1).15 Sufficient investment, training and practical support (n=1).15 Sufficient investment, training and practical support (n=1).15 Challenges with documentation quality (e.g., in relation to accuracy, detail, length and/or structure of AVT outputs) (n=7).15 31 36 40 45 47 48 Challenges with documentation quality (e.g., in relation to accuracy, detail, length and/or structure of AVT outputs) (n=7).15 31 36 40 45 47 48 Varied effectiveness across different settings, types of consultations (n=4)44 48–50 or scribes (n=1).44 Varied effectiveness across different settings, types of consultations (n=4)44 48–50 or scribes (n=1).44 Lack of integration with current workflow and/or existing documentation processes (n=3).15 48 49 Lack of integration with current workflow and/or existing documentation processes (n=3).15 48 49 Technical challenges with accessing and using the tools (n=4).15 40 44 48 Technical challenges with accessing and using the tools (n=4).15 40 44 48 Concerns about data privacy (n=2).15 47 Concerns about data privacy (n=2).15 47 Lack of familiarity with AVT/readiness for innovation (n=2).15 48 Lack of familiarity with AVT/readiness for innovation (n=2).15 48 AVT tools not impacting documentation time (i.e., clinicians feeling they could type the notes more quickly than receiving and editing an AVT output) (n=2).31 49 AVT tools not impacting documentation time (i.e., clinicians feeling they could type the notes more quickly than receiving and editing an AVT output) (n=2).31 49 Lack of personalised template designs which align reality with AVT output (n=2).15 47 Lack of personalised template designs which align reality with AVT output (n=2).15 47 Undefined approaches and limited consistency when evaluating the accuracy and safety of AVT in practice (n=1).48 Undefined approaches and limited consistency when evaluating the accuracy and safety of AVT in practice (n=1).48 Environmental factors (e.g., noise) and tool reliability (n=1).15 Environmental factors (e.g., noise) and tool reliability (n=1).15 AVT implementation could result in clinicians being expected to increase patient volume (n=2).15 31 AVT implementation could result in clinicians being expected to increase patient volume (n=2).15 31 AVT, ambient voice technology; EHR, electronic health record. Consultation context AVT was particularly useful when consultations were structured or less complex (n=3),31 40 50 for generating narrative summaries of consultations (n=2),15 40 supporting documentation accuracy (n=2)15 41 and enhancing accountability/responsibility (particularly in complex settings) (n=1).15 One study reported that AVT was notably useful for neurodivergent clinicians (e.g., with dyslexia) when managing documentation tasks.15 AVT was perceived as less effective when used for consultations with non-English speaking patients (n=3),15 40 48 patients with strong accents or fragmented speech (n=1)15 and in more complex workflows or unstructured settings requiring real-time adaptability (n=6)15 33 44 45 47 50 (e.g., mental health and/or psychiatry settings, during physical examinations). Two studies reported positive experiences when AVT was used with interpreters.15 44 Critical appraisal All MMAT quality assessment ratings are included in supplementary file 3. Qualitative studies (n=2) met the required criteria and were of high quality. All mixed methods studies were of moderate, low or very low quality (n=2, n=2 and n=4, respectively). Of the 11 quantitative studies, none were of high quality. Four were of good or moderate quality (n=3 and n=1, respectively), with the remaining rated low or very low quality (n=2 and n=5, respectively). Across the outcome domains assessed, no observable differences based on MMAT quality ratings were identified in relation to the nature of impact or type of literature source. Reported declarations of competing/conflicts of interest are summarised in supplementary file 2. Authors in nine papers declared conflicts,33 38–43 45 50 thus the findings regarding potential benefits of AVT should be considered in this context. Discussion Key findings Review findings suggest that AVT may be associated with reductions in documentation burden and improvements in staff well-being and perceived care quality when used for documenting consultations in real-world practice. However, evidence was mixed regarding the quality of AVT outputs and very few studies explored the impact of AVT on wider pathway productivity, patient safety and cost. Although there were positive experiences of using AVT, these varied depending on a range of influencing factors (e.g., EHR integration, variation in scribe quality, poor documentation quality and data transparency) and the clinical context in which AVT was used (e.g., more effective in structured vs complex consultations and settings). The majority of studies were short-term pilots, variable in methodological quality, with notable heterogeneity in the approaches to evaluating impact. How findings relate to previous research Our findings align with existing reviews2 17 20 and the rationale for increased interest in AVT and subsequent rapid implementation.2 3 11 17 For example, AVT use can help to improve documentation efficiency and reduce administrative burden for clinicians.2 3 11 17 The use of AVT may also enable consultations to be more patient-focused.2 3 11 17 Concerns relating to documentation quality and the risks associated with AVT (e.g., including incorrect details in notes and/or hallucinating diagnoses) have been previously documented.2 17 These inaccuracies have notable risks for patient safety and misdiagnoses when AVT is used in real-world practice and beyond theoretical testing.2 Therefore, although AVT may lead to improvements in documentation efficiency, the need for quality assurance was evident; highlighting the importance of clinician accountability and monitoring of use to ensure practice is safe and ethical.2 15 This may have implications for any efficiency gains, as reviewed studies highlighted that substantial editing could offset any associated improvements.15 31 33 40 45 Findings further emphasise the importance of understanding the risks and unintended consequences of using AI, alongside the potential benefits often anticipated.8–10 Findings also demonstrated that the consultation context and clinical setting impacted the effectiveness and usability of AVT. This supports existing research which has suggested that AVT may be less effective in certain contexts, particularly those which are more complex rather than structured (e.g., mental health/psychiatry) and for consultations with non-English speaking patients.2 20 The current evidence strengthens the rationale for understanding more clearly when AVT works well and when it may not be appropriate to use, noted in previous literature.2 15 20 Factors influencing adoption, such as workflow integration (notably with EHRs for AVT), technical challenges, leaders/champions to support implementation, familiarity and readiness, transparency and effective training, reflect existing work related to the usability and acceptability of digital innovations in healthcare contexts.1 20 51 52 Advancing previous reviews, our findings outline the heterogeneous approaches used to measure and evaluate the impact of AVT, which primarily focused on the impact for staff (i.e., documentation burden, documentation quality, well-being). Though care quality was studied, this mainly relied on self-reported experiences, with few studies exploring patient experiences. We highlight the paucity of evidence evaluating how AVT can impact wider productivity in the healthcare system, beyond more immediate impacts during or following consultations. For example, how saving documentation minutes may release clinical capacity and the impact of these gains. Such evidence is essential, as using digital innovations to improve health service efficiency is at the centre of current policy.10 Evidence relating to cost and patient safety are notably limited, both of which are important outcomes to consider when implementing an innovation, particularly within a national health service.16 53 Quantitative evaluations of the impact of AVT across the outcomes reported and any causal inferences that may be drawn should also be considered in the context of short-term observational studies of variable design and in differing healthcare settings, often lacking a comparative analysis which would permit an assessment of impact magnitude. As we do not yet know enough about the long-term and sustained impact of AVT, review findings suggest that there is currently limited evidence to support widespread clinical implementation. However, this reflects how digital innovations in health care1 and AVT specifically7 46 are evolving areas of practice and research, with evaluations being conducted rapidly in response to current implementation pace. With already limited evidence, it should be noted that competing/conflicts of interest were reported by some authors. While not negating their conclusions, the relevant findings should be considered in this context. Our findings speak to a wider challenge when evaluating digital innovations—robust, independent evidence is needed to inform adoption while adoption is needed to generate real-world evidence.1 53 Varied approaches, heterogeneous quantitative measures and limited assessment of longer-term impact may reflect the rapidity of implementation, coupled with an absence of clear frameworks to guide evaluations of AVT.16 48 49 Strengths and limitations This review provides a focused synthesis of empirical evaluations of AVT use in routine clinical practice and examines both quantified and perceived impacts across staff, patient and system domains. By detailing how outcomes were defined and measured, it highlights substantial heterogeneity with implications for future evaluation design. The predominance of US-based studies may limit applicability to other health systems. As a rapid review, certain methodological decisions (e.g., including single-reviewer screening with partial verification) may increase the risk of missed studies in a rapidly evolving field. Therefore, the review may not have captured all available evidence and longer-term studies/outcomes. Although the search strategy was iteratively developed based on pilot searching and librarian support, the inclusion of some products specifically (i.e., DAX) may have influenced search results. The variation in terminology used to describe AVT (e.g., AI-scribe, ambient scribes and/or digital scribes) may have also influenced the risk of missing studies. Such variation and often limited detail regarding functionality made the screening process more complex, often requiring discussions between multiple reviewers to determine paper eligibility. Implications for policy and practice Although AVT is increasingly being implemented in clinical settings, the current evidence base does not yet demonstrate consistent or generalisable improvements in clinical outcomes, system productivity or cost effectiveness. Decisions regarding adoption should therefore consider the limited and evolving nature of the evidence base. This mirrors patterns seen in evaluations of AI across healthcare, where early adoption often precedes robust effectiveness data.1 Implementation ahead of robust evaluation risks overestimating benefits and under-recognising potential unintended consequences, including impacts on workflow, equity and patient safety. Importantly, rapid adoption and regulatory compliance should not be interpreted as evidence of effectiveness, particularly given the heterogeneity of AVT products and evaluation approaches. Given the geographical skew of the current evidence indicated in review findings, policy decisions regarding implementation should also interpret these findings with caution; reflecting on differences in health systems across countries. Future research Future research should prioritise longer-term evaluations permitting longer-term outcomes, including clearly defined comparators, standardised and transparent outcome measures, independent economic analyses incorporating full implementation costs and formal safety monitoring frameworks. The complexity of integrating AI into established clinical workflows highlighted in previous work on digital documentation systems,20 54 underscores the importance of rigorous implementation and evaluation design. However, structured and comprehensive data collection is needed to enable quantitative and economic analyses. Challenges related to language diversity, performance in non-English-speaking consultations and more complex settings,2 20 further emphasise the need to examine equity impacts and performance across diverse patient groups, to understand when AVT is more or less effective. Greater inclusion of patient-reported and qualitative data is also needed, given the relative scarcity of in-depth qualitative studies which include patient experience.40 Additional qualitative work would enable further exploration of staff and patient experiences, including how the impact of AVT on documentation burden and well-being may translate into efficiency gains and any unintended consequences of AVT use. The development of shared evaluation frameworks and reporting standards may support greater methodological consistency and comparability as implementation evolves.16 48 49 Future reviews and meta-analyses, if feasible, may explore in greater depth factors which influence quantitative outcomes (e.g., documentation burden) following AVT implementation. Conclusions Current evidence of AVT impact in real-world clinical practice remains limited. While some studies suggest potential reductions in documentation burden and improved staff well-being and perceived care quality, findings are predominantly based on short-term observational studies and heterogeneous outcome measures. Evidence relating to wider system productivity, safety, cost effectiveness and equity is sparse. Robust, independent and longer-term evaluations using standardised outcome measures (e.g., pyjama time/after-hours documentation time, Maslach Burnout Inventory, NASA Task Load Index, PDRQ-9), supported by complementary qualitative methods, are required before definitive conclusions about effectiveness, safety and system-level value can be drawn.