Protocol for a Scoping Review of Digital Scribes in Primary Care: Design, Implementation, Accuracy and Impact

Background

Documentation in General Practice

General practice is a complex and high-pressure setting, marked by increasing multimorbidity, shorter consultation times, and expanding administrative responsibilities¹. Clinicians are required to manage diverse presentations and deliver patient-centred care, while simultaneously maintaining comprehensive documentation within electronic health records (EHRs). Although EHRs have improved legibility, continuity, and data availability, they have also contributed to a substantial increase in clinician workload². Documentation now consumes a disproportionate share of clinical time, with general practitioners (GPs) reporting an average of 2–3 hours per day spent on administrative tasks, often beyond scheduled clinic hours^3,4.

This growing burden is linked to adverse effects on clinician wellbeing and professional satisfaction. Burnout, emotional exhaustion, and reduced job satisfaction have been consistently associated with EHR-related tasks⁵. The psychological impact is compounded by a sense of reduced control over time and diminished capacity for meaningful patient interaction. Although strategies such as time-blocking and delegation have been explored, these are often insufficient without structural changes at the system level³. Furthermore, there is no standardised framework for measuring documentation burden, limiting the comparability of interventions across settings⁶.

In this context, the need for effective, scalable solutions to reduce documentation workload and preserve the quality of care has become a priority in primary care research and policy.

Promise of Digital Scribes

One proposed solution is the use of digital scribes—automated systems that employ speech recognition and natural language processing (NLP) to transcribe and structure clinical conversations into documentation. Digital scribes operate in real time or asynchronously, aiming to reduce clinician time spent on notetaking while enhancing documentation quality⁷. These technologies have progressed from simple transcription tools to more sophisticated models capable of generating structured SOAP (Subjective, Objective, Assessment, Plan) notes and integrating with EHRs^8,9.

Early studies suggest digital scribes may improve workflow efficiency and free up time for patient care. For example, a recent evaluation found that digital scribes reduced documentation time without compromising content quality, though user experience varied significantly depending on local context and clinician preference¹⁰. In some settings, digital scribes also enabled broader roles for other team members, contributing to more collaborative models of care⁹.

Despite this promise, implementation remains limited in primary care, partly due to concerns around technical integration, start-up costs, data privacy, and system usability⁸. Existing evaluations are often pilot studies or small-scale observational reports, many of which are situated in secondary care contexts. As a result, there is limited generalisability to the unique workflows and constraints of general practice.

Recent Advances in Digital Scribes

Recent advances in artificial intelligence, particularly the development of large language models (LLMs) such as GPT-4, have significantly expanded the capabilities of digital scribes. These models can generate detailed clinical notes, summarise longitudinal patient histories, and extract key concepts from unstructured data^11,12. Their application has the potential to enhance both documentation and clinical reasoning processes by embedding context-sensitive language understanding into automated outputs.

However, the application of LLMs in clinical documentation is not without limitations. Errors such as misinterpretation of speech, factual inaccuracies (i.e. "hallucinations"), and variable comprehensiveness have been documented¹³. Furthermore, these models often lack specific training on clinical datasets, which can compromise their reliability in real-world use. Challenges with reproducibility, bias in language outputs (e.g. under-representation of certain populations), and inconsistencies in summarisation raise legitimate concerns about patient safety and equity^14,15.

While LLM-powered scribes hold theoretical promise, there is limited evidence regarding their actual performance in primary care. Most recent studies are technology-focused and do not assess clinical integration, impact on outcomes, or unintended consequences such as automation bias or loss of clinician vigilance.

Risks and Limitations of AI-Generated Documentation

AI-generated documentation, while efficient, introduces a distinct set of risks. Hallucinations—or fabricated content presented as fact—pose a clear danger to clinical accuracy and patient safety. In radiology, for instance, false AI-generated statements have necessitated the development of hallucination detection tools like ReXTrust^16,17. In primary care, where patient histories are central to decision-making and continuity of care, such errors could have more diffuse but equally serious effects.

Bias is another critical concern. Generative models may inadvertently reinforce existing disparities through underrepresentation or stereotyping in their training data. Studies have found gender and ethnicity-based distortions in AI-generated medical documentation, particularly in fields such as nuclear medicine¹⁴. Privacy concerns also persist, particularly in ambient recording contexts where data is passively collected and analysed. Without robust governance frameworks, the potential for misuse or breach of sensitive information remains high¹⁸.

Addressing these risks requires not only technical improvements but also clear guidelines on appropriate use, informed consent, data handling, and clinical oversight.

Ethical and Regulatory Considerations in Primary Care

The ethical and legal implications of ambient documentation tools are particularly complex in general practice, where longitudinal relationships, holistic care, and continuity are foundational. GPs routinely handle highly sensitive information, and breaches of confidentiality—whether real or perceived—can undermine trust and engagement^19,20.

While guidance from regulatory bodies such as the General Medical Council highlights the importance of maintaining patient confidentiality, there is limited clarity on how emerging technologies like digital scribes fit within existing frameworks²⁰. Issues such as data lifecycle management, consent procedures, and medico-legal accountability require careful consideration. Liaw et al. (2020)¹⁹ argue for integrating data governance with quality management, noting that ethical data curation must be ongoing and responsive to system evolution²¹.

Moreover, the adoption of digital scribes intersects with broader shifts in primary care towards digital transformation and patient-centred models. This necessitates a nuanced understanding of how such tools affect clinician–patient communication, power dynamics, and trust.

Current Evidence and the Need for an Updated Synthesis

Despite the growing interest in digital scribes, there has been no comprehensive synthesis of their application in primary care since 2021. Existing reviews have largely focused on early-stage technologies or hospital-based use, providing limited insight into general practice settings^7,8.

Recent work on human medical scribes has demonstrated benefits such as improved clinician satisfaction, increased documentation quality, and enhanced patient interactions^2,22,23. However, the extent to which digital scribes can replicate or exceed these benefits remains unclear. Preliminary studies have shown reductions in documentation time and improvements in provider workflow, but robust data on clinical outcomes, diagnostic accuracy, and patient safety is lacking²⁴.

Given the rapid development of LLMs and their increasing availability in clinical settings, a timely review is needed to understand the landscape of digital scribes in primary care. Such a synthesis can inform the development of future tools, guid regulatory frameworks, and ensure that innovation aligns with the realities and values of general practice.

Aim and objectives

This scoping review aims to systematically map the literature on digital scribe technologies developed for, or with potential for use in, primary care settings. The specific objectives are:

1.   To map the scope and nature of published literature on digital scribes in primary care, including study design, publication characteristics, geographic distribution, and identification of evidence gaps.

2.   To identify digital scribe technologies described in the literature and characterise each by technical features, intended clinical use, target users, and implementation approach.

3.   To synthesise evidence on the technical performance of digital scribes in real-world and simulated primary care settings, focusing on documentation accuracy, coding practices, and integration with existing clinical systems.

4.   To synthesise evidence on the clinical impact of digital scribes in primary care, particularly their effects on workflow efficiency, clinician experience, and patient care outcomes.

5.   To document regulatory, legal, and ethical considerations associated with digital scribe tools, including data governance, consent processes, and medico-legal implications, as reported in the literature.

Reporting and registration

This scoping review will be conducted in accordance with the Arksey and O'Malley framework, enhanced by the methodological advancements proposed by Levac et al. (2010)²⁵, and reported according to the PRISMA-ScR checklist²⁶. The protocol will be reported according to the PRISMA-P guidelines for protocol development²⁷. This published protocol serves as the formal record of the planned methodology and will function as a substitute for prospective registration. Supplementary materials such as the detailed search strategy and data extraction forms will be made available via Open Science Framework (OSF)²⁸.

Eligibility criteria

This scoping review applies a modified Population–Concept–Context (PCC) framework, as recommended by the Joanna Briggs Institute²⁹. Given the technological focus of the review, the “Population” and “Context” domains are considered together and interpreted broadly. Here, both refer broadly to primary care and comparable settings.

Concept: Digital Scribes

For the purposes of this review, digital scribes are defined as technologies that automatically convert speech into text using machine-based methods, such as automatic speech recognition (ASR) or natural language processing (NLP). Tools that rely on human transcription, whether performed live (in person or remotely) or retrospectively from recordings, are excluded.

To be eligible, the output of the system must contribute directly to the creation of a medical or paramedical document—such as a clinical note, patient summary, report, or structured record. Tools that transcribe speech for non-documentation purposes (e.g., voice navigation, command recognition, or general virtual assistants) are not considered digital scribes for the purposes of this review.

Context: Primary Care

Eligible studies must demonstrate relevance to primary care, either through direct implementation in primary care or evaluation in comparable outpatient, community-based, or paraclinical contexts. Studies conducted exclusively in inpatient, specialist, or tertiary care settings will be excluded unless a clear rationale is provided for their applicability to general practice.

Additional Criteria

6.   Language: Only studies published in English will be included.

7.   Study Design: All study designs will be considered, including qualitative, quantitative, mixed-methods, technical, and implementation-focused research.

8.   Source Type: Editorials, opinion pieces, preprints, conference abstracts, dissertations, and other forms of grey literature will be included if they offer substantive insight into digital scribes and their relevance to primary care.

Information sources

A wide range of sources will be searched to ensure a comprehensive overview of the literature on digital scribes in primary care, specifically:

9.   Bibliographic Databases — MEDLINE (via Ovid), Embase, CINAHL (via EBSCO), Scopus, Web of Science Core Collection

10.   Preprint Servers — medRxiv, bioRxiv, arXiv,

11.   Dissertation and Thesis Databases — ProQuest Dissertations and Theses Global

12.   Grey Literature Sources — OpenGrey, National Institutes of Health (NIH) RePORTER, Agency for Healthcare Research and Quality (AHRQ) Reports, National Health Service (NHS) Digital Publications

13.   Systematic Review Databases and Registries — Cochrane Database of Systematic Reviews, JBI Evidence Synthesis, PROSPERO (International Prospective Register of Systematic Reviews)

14.   Clinical Trial Registries — ClinicalTrials.gov, EU Clinical Trials Register, ISRCTN Registry, WHO International Clinical Trials Registry Platform (ICTRP)

Search strategy

This review aims to conduct a comprehensive (i.e. sensitive) search of the literature on digital scribes. However, to support efficient screening on large bibliographic databases, the search strategy is structured to prioritise specificity early in the process. The search is divided into six non-overlapping segments that allow us to screen the most relevant records first, while still capturing broader or less explicitly labelled studies. For smaller databases, a simplified search approach will be used.

The six searches are organised along two dimensions: (a) whether the study includes a reference to primary care, and (b) how directly the search terms refer to digital scribes, using a three-tiered conceptual hierarchy:

Scribe 1: Studies that use the term “scribe” explicitly.

Scribe 2: Studies that use synonyms or synonymous terms (e.g., “voice technology,” “ambient AI”) to describe technologies with similar functions.

Scribe 3: Studies that focus on underlying technologies (e.g., “automatic speech recognition,” “natural language processing”) that may be used in digital scribes but are not specific to them.

Boolean “NOT” operators are used to ensure that each segment captures only unique records not already retrieved in earlier searches. This mutually exclusive structure enables a progressive screening process, beginning with the highest-specificity records and extending to more general or foundational studies that may still be relevant. Table 1 summarises the logic and expectations for each segment.

Selection of sources of evidence

All search results will be imported into Covidence for deduplication and screening. Screening will proceed in two stages: title and abstract screening, followed by full-text review. Both stages will be conducted independently by two reviewers, with disagreements resolved through discussion or adjudication by a third reviewer. Reasons for exclusion at the full-text stage will be recorded and reported in a PRISMA-ScR flow diagram. Table 2 describes a structured screening algorithm that will guide the selection process, ensuring consistent application of the eligibility criteria.

Data collection

Data will be extracted and organised across four interlinked tables: “Papers”, “Tools”, “Tech-Legal”, and “Fidelity-Clinical”. This structure reflects the hierarchical nature of the data pertaining to digital scribe tools. Individual studies may describe multiple tools, and individual tools may be used or evaluated across multiple studies. A single extraction table would be insufficient to accurately capture these many-to-one and one-to-many relationships; therefore, separate but linked tables are used to preserve the relational structure of the data.

Papers table

The Papers table will capture bibliographic details for each included source, including author, year, country, and publication type. This table provides a comprehensive overview of the studies included in the review and serves as the foundational dataset to which all other tables will link. [Paper Table]

Tools table

The Tools table will catalogue the digital scribes identified in the literature, linking each tool to one or more associated papers [Tools Table]. This table will include data on tool characteristics, such as name, developer, and reported functionalities. Each entry in the Tools table will be connected to the relevant study or studies in the Papers table.

Tech-Legal table

The Tech-Legal table will capture two distinct types of data: technical data and legal data (including ethical and regulatory considerations) [Tech-Legal Table]. Technical data (e.g., software architecture, data storage protocols) will be linked to specific tool-paper pairs, as technical specifications typically pertain to the functionality and implementation of individual digital scribe tools. Legal data (e.g., data protection frameworks, consent processes, regulatory compliance) will be linked to the associated paper rather than specific tools, as these considerations often reflect broader legal or ethical contexts that apply across multiple tools or general implementation settings. This structure ensures that both tool-specific technical data and broader legal data are systematically documented, preserving the relational integrity of the data without conflating distinct data types.

Fidelity-Clinical table

The Fidelity-Clinical table will capture fidelity outcomes (e.g., transcription accuracy against gold standards) and clinical impacts (e.g., effects on documentation quality, workflow efficiency, clinician or patient experience) reported in relation to specific tool-paper pairs [Fidelity-Clinical Table].

Data extraction will be conducted independently by two reviewers using piloted forms, with discrepancies resolved through consensus among the review team, with consultation from a third reviewer if necessary. Data will be managed using Microsoft Excel, and relevant supplementary materials will be made available via OSF to ensure transparency and reproducibility. The extraction framework may be iteratively refined as familiarity with the included studies develops, in line with standard scoping review methodology. Links to the initial data charting tables are provided under each heading; however, these tables are subject to revision based on findings in the literature. Any modifications to the structure, content, or focus of the tables will be transparently reported.

Data management

All search results will be imported into Covidence for deduplication, screening, and preliminary data extraction. Following screening, extracted data will be exported into Microsoft Excel for coding, organisation, and synthesis. Extracted data will be organised in accordance with above data collection plan, and then terminology will be standardised to facilitate comparison. Ambiguities and inconsistencies will be resolved through consensus among the review team, with consultation from a third reviewer if necessary. An audit trail will document any modifications to the extraction framework.

Critical appraisal

Although formal quality assessment is not a requirement for scoping reviews, an optional critical appraisal will be conducted for interventional and observational studies to assess the robustness of their findings. The ROBINS-I tool will be used for observational studies³⁰, and the RoB 2 tool will be applied to randomised trials³¹. Findings from these assessments will not exclude studies but will be used to contextualise the results in terms of study quality.

Data synthesis

Data synthesis will be organised across five distinct categories, corresponding to the review’s research questions and data charting framework. In each category, findings will be synthesised using tabulation and narrative summary.

15.   Bibliometric summary of the included studies

16.   Design and implementation details of eligible digital scribe tools

17.   Technical performance and transcription fidelity

18.   Impact on clinical workflow and clinical outcomes

19.   Regulatory, legal, and ethical considerations

This structured yet broad synthesis will produce a comprehensive high-level mapping of the evidence on the use and potential use of digital scribe technologies in primary care.

Summary

This scoping review aims to systematically map the existing literature on digital scribe technologies in primary care and comparable settings. It will provide a comprehensive overview of published studies, focusing on the design, implementation status, accuracy, and clinical impact of identified tools. By organising these findings within a structured framework, the review will identify gaps in evidence and areas for further research, clarify how digital scribes are currently integrated into clinical documentation workflows, and provide a practical reference for those considering the adoption or evaluation of digital scribe technologies in primary care.

Strengths

Several methodological strengths support the rigour and relevance of this review. The use of a tiered, mutually exclusive search strategy balances sensitivity with screening efficiency, ensuring relevant studies are identified while managing resource demands. The relational data extraction framework, structured across multiple interlinked tables, preserves methodological nuance and allows for more comprehensive analysis of complex data relationships and facilitates richer analysis. Additionally, inclusion of diverse study types and grey literature ensures that the review captures both peer-reviewed and practice-relevant insights, broadening its applicability to clinical and policy audiences.

Weaknesses

Methodological limitations include the exclusion of non-English-language sources, limiting global generalisability, and the inability to systematically search the grey literature, increasing the risk of missing less well-documented tools. Secondary care studies will be included but subjectively assessed for their relevance to primary care, introducing potential bias.

Certain limitations arise from the nature of scoping reviews and the defined scope of this review. Rapid technological evolution may render some tools outdated before the review is completed, potentially affecting the relevance of findings. The focus on fully or primarily automated systems excludes hybrid models, which may still hold relevance in settings transitioning towards full automation. Additionally, as a scoping review, the aim is to map the literature rather than critically appraise study quality or synthesise findings quantitatively, meaning that the review will describe the breadth and scope of the evidence rather than evaluate the strength of individual studies.

Implications

The findings of this review will provide an overview of the current state of digital scribe technologies, including tool design, implementation status, accuracy, and clinical impact, offering a practical reference for developers, researchers, clinicians, and policymakers.

For developers, the review will map existing tools and their technical specifications, identifying common design features, functional gaps, and areas for enhancement, particularly around accuracy and workflow integration. This may inform iterative design and targeted development in line with emerging clinical needs. For health service researchers, the synthesis will highlight underexplored technologies and contextual factors affecting implementation, outlining priority areas for empirical investigation. The structured framework developed in the review may also serve as a foundation for systematic reviews or meta-analyses of clinical impact. For clinicians, the review will summarise reported clinical impacts, including documentation accuracy, workflow efficiency, and integration challenges. This information can support decision-making around whether and how to incorporate digital scribes into practice, particularly in primary care settings. For policymakers, the review will outline regulatory and legal considerations related to data governance, consent, and medico-legal responsibility, clarifying how existing frameworks address the use of digital scribes in primary care. This may inform policy development, particularly as hybrid and semi-automated models gain traction.