Guided by a pragmatic paradigm, emphasizing practical inquiry and the use of diverse methods to address real-world problems, this study used a convergent mixed-methods design. Survey and interview data were collected in parallel to examine how the AI scribe was integrated into clinical practice [27,28]. The survey findings offered valuable descriptive and contextual information, while the interviews provided richer, more detailed insights into participants’ experiences. Bringing these approaches together strengthened the credibility of the results and allowed for a more holistic understanding of the factors influencing acceptance [29]. A qualitative-dominant mixed-methods design was used (QUAL+quan), with qualitative data forming the primary strand and quantitative data used to complement and contextualize the findings [30]. The reporting of this study was guided by the MMR-RHS (Mixed Methods Reporting in Rehabilitation & Health Sciences) guidelines (Checklist 1).

The study took place within a large Australian public hospital and health service located in a metropolitan region. This service employs around 1600 allied health professionals and delivers hospital and community services to a population exceeding 700,000 [31].

The research was undertaken by a multidisciplinary team that included hospital consumers, allied health clinicians, clinical leaders, researchers, and a representative from the hospital’s digital health unit, ensuring the study’s relevance and credibility. Consumer representatives were engaged through the health service’s Consumer Advisory Group (HS and EJM). One brought prior research experience, while the other was new to the process; both were compensated for their contributions. They were involved throughout the whole research life cycle and are named as coauthors. To further strengthen rigor, a reference group comprising consumers, clinicians, and health service and research leaders was convened to provide guidance and oversight at critical stages of the project. The interviews and analysis were conducted by an experienced hospital social worker and health services researcher with no strong pre-existing views on the AI scribe beyond the need for further study. Most participants were not previously known to the researchers; some were known professionally, but there were no close or personal relationships.

Ethical approval was obtained from Gold Coast Hospital and Health Service Human Research Ethics Committee (HREC/2025/QGC/118766). Research governance approval was given by the Gold Coast Hospital and Health Service (SSA/2025/QGC/118766). Participants were provided with an electronic information sheet via REDCap (Vanderbilt University) outlining the study, including the risks and benefits of participation. Participation was voluntary, and informed consent was obtained electronically through REDCap. No compensation was provided to the clinicians who participated in this study. All study data were deidentified for analysis. However, survey data for participants who provided their email addresses were identifiable to two of the investigators, who had access to the raw data extracted from REDCap, prior to analyses.

Eligible participants were allied health clinicians from professions including physiotherapy, occupational therapy, social work, dietetics, speech pathology, pharmacy, psychology, audiology, and podiatry with over 2 years of clinical experience. The decision to exclude participants with less than 2 years of experience was made because some would have been ineligible to use the AI scribe due to permissions set by some areas of the health service, and because the study aimed to capture the perspectives of more established clinicians. Although HP6 and HP7 classifications were included as response options in the survey demographics to enable participation from clinicians in senior leadership roles, no participants were recruited from these grades. At the commencement of data collection, approximately 280 allied health clinicians had been granted access to the AI scribe. A convenience sampling approach was used for the clinician survey, while purposive sampling guided the recruitment of interviewees to ensure representation across professions. Both adopters and nonadopters were included to provide a comprehensive understanding of the factors influencing adoption. Invitations to participate were distributed via email, team meetings, and targeted outreach informed by AI scribe usage data.

Quantitative and qualitative data collection was undertaken concurrently between August 2025 and October 2025.

Survey data were exported from REDCap into Microsoft Excel. Descriptive statistics (frequencies and percentages) were used to summarize closed-ended items. Responses from the 5-point Likert scale were dichotomized for analysis. “Agree” and “Strongly Agree” were combined to represent overall agreement, while “Disagree” and “Strongly Disagree” were combined to represent overall disagreement, and percentages were reported accordingly. Neutral responses were not grouped with agreement or disagreement responses but were instead treated as a separate category. Survey items, including reverse-worded statements, were analyzed using their original coding, with no scale-level reverse scoring applied, as responses were used for descriptive and comparative analyses alongside qualitative findings.

Interview transcripts were coded and themed in NVivo (Lumivero), with a deductive approach guided by COM-B to identify determinants of behavior [25,33]. Initial coding was completed by 1 researcher (LR), who then grouped codes into potential themes aligned with COM-B domains. These themes were reviewed and refined through discussion with the wider research team to resolve discrepancies and ensure interpretive rigor. Finally, these were considered in relation to the behavior change wheel to identify potential intervention functions and policy categories relevant to supporting AI scribe adoption [25]. A pragmatic approach was taken, recognizing that some strategies identified by participants did not align neatly with the ehavior change wheel and were therefore incorporated flexibly to ensure practical applicability.

Data synthesis adopted a qualitatively driven mixed-methods design, prioritizing qualitative findings for their depth and contextual richness relative to survey data. Interviews offered detailed insights into experiences and perceptions, while survey data provided complementary breadth and supported triangulation of key themes. Qualitative data from interviews were first analyzed to establish a foundational framework of themes. Subsequently, quantitative survey data were analyzed and mapped against these emergent qualitative categories. This mapping process confirmed a high degree of alignment between the 2 datasets, allowing for the integration of descriptive statistics to validate and illustrate the prevalence of the qualitatively identified themes. This approach reflects best practice for qualitatively led mixed-methods research, where qualitative evidence serves as the primary interpretive lens [30]. Open-ended survey responses were reviewed alongside interview data to check for consistency; however, they did not generate additional unique insights and are therefore not presented here.

The study included 27 clinician interviews across multiple allied health disciplines and 110 survey submissions. Of these, 97 surveys were analyzed. Eleven were excluded due to substantial incompleteness (consent or demographic data only), and two were excluded for not meeting the eligibility criteria (not allied health clinicians). To ensure a comprehensive representation of perspectives, partially completed surveys were retained. Of the 97 surveys analyzed, 58 were completed by adopters of the AI scribe, and 39 were nonadopters. For the interviews, 22 participants were adopters, and 5 were nonadopters. Further demographic characteristics of participants are summarized in Table 1.

Analysis revealed an interchange among COM-B domains that shapes adoption and integration into practice. Qualitative data from clinician interviews are presented as the primary narrative, and survey findings are integrated throughout to reinforce and elaborate on key themes. Figure 1 outlines the key determinants, while Table 2 summarizes these under the COM-B components, indicating whether each acted as a barrier or enabler. Findings are reported in relation to the COM-B model. Potential intervention strategies derived from the behavior change wheel mapping are presented in the “Discussion” section [25].

This study examined the factors influencing the adoption and integration of an AI scribe among allied health clinicians in an Australian public health service, utilizing the COM-B model to guide the analysis. Findings revealed a delicate interplay between COM-B domains, with multiple determinants acting as barriers, enablers or both to implementation. Although research on AI scribe implementation remains limited, some findings align with existing literature—for example, the need for dedicated time to support adoption and training—which also emerged as an important finding in this study [24].

Building on earlier research exploring clinician perceptions of AI implementation [26], our study provides more nuanced insights specific to the implementation of AI scribe technology. While similar themes emerged, such as the need for leadership support, time and capacity constraints, and targeted education and training, unlike in Hoffman et al [26], concern about role replacement was not noted as a barrier to use, and apprehension about AI in health care was only a minor barrier. Instead, adoption appeared to be more strongly influenced by practical challenges associated with integrating the AI scribe into routine practice, such as having the necessary time, means (eg, suitable equipment), knowledge, skills, support, and organizational endorsement to initiate its use. This may reflect that most participants in our study were already adopters of the AI scribe. Alternatively, it may be because many participants recognized its potential usefulness and were instead grappling with its integration into practice. This highlights that, while broader understandings of AI in health care provide an important backdrop, more nuanced studies are needed to examine how specific technologies are implemented within real-world contexts [12].

Concerns regarding the processes of obtaining patient consent during encounters using the AI scribe, including the time burden, have been previously identified within the hospital outpatient setting, with clinicians identifying competing priorities during short appointment times as a challenge [34]. Lawrence et al [34 p.11] suggest shifting preliminary consent-related conversations to “earlier touchpoints in the patient journey,” such as at a reception desk or via a digitally enabled patient portal, and then having a secondary consent conversation with clinicians regarding consent [34]. Further exploration of these strategies and others that help obtain true informed consent when using AI scribes is important to minimize any ethical risk of assumed or poorly understood consent.

Findings were mapped to the behavior change wheel to identify the intervention functions most relevant to supporting AI scribe adoption, as shown in Table 3. This process enabled the translation of behavioral determinants into actionable implementation strategies, highlighting areas where targeted support may facilitate uptake and sustained use (Textbox 1).

Together, these findings highlight that successful AI scribe implementation in allied health may depend on practical strategies that prioritize both Capability and Opportunity through reliable infrastructure, dedicated time, and access to tailored training and support, while understanding the underlying individual motivational factors that influence adoption, including the burden of traditional documentation practices, prior digital experience and perceptions of AI safety and reliability. Thus, effective implementation of AI scribe requires addressing both system-level factors (eg, training, infrastructure, and resourcing) while also acknowledging and supporting individual factors and attitudes across different experiences and perceptions that shape adoption and use. For example, a greater proportion of nonadopters had concerns related to privacy compared to those who had adopted the AI scribe. As such, providing targeted education to clinicians may alleviate some of these concerns, thereby addressing this barrier to adoption. Fostering supportive social, organizational, and emotional environments may also be critical for engaging nonadopters and promoting sustained, confident use. Addressing these tangible requirements can provide a strong foundation for effective and enduring adoption of AI scribes across diverse allied health settings.

As a qualitatively dominant mixed-methods study, interviews provided depth and nuanced insight into clinicians’ and patients’ experiences with the AI scribe, while the survey offered broader contextual understanding and supported the wider relevance of emerging themes. A key strength of the study was the diversity of allied health clinicians involved, which enabled the capture of varied perspectives and enhanced the applicability of the findings.

Some limitations were also evident. Although the survey included clinicians from multiple disciplines, the number of respondents was small relative to the approximately 1600 allied health staff employed by a single Australian health service. While demographic comparisons were not completed, the findings may not fully reflect the wider workforce. Further research is also needed to better understand the perspectives and experiences of staff with less than two years of experience, as these were not a focus of the current study. Future research may also wish to recruit from multiple sites in different geographical areas to enhance the generalization of findings.

A further limitation was the use of a bespoke survey instrument that did not undergo formal psychometric validation. While the survey was theoretically informed by the COM-B and piloted with staff to maximize content validity, including a more formal validation of survey tools in future research may increase reliability and generalizability of the findings. Also, the survey design required respondents who had not completed AI scribe training to provide a response to a training-related item, with a neutral option available to reflect their lack of experience. However, as no untrained participants reported using the AI scribe with patients, this is unlikely to have influenced the study findings. Finally, while rigor was ensured through reflexive engagement with the data and regular discussion with the research team, we acknowledge that the use of multiple coders may have further enhanced the credibility of the qualitative analyses undertaken.

This study advances understanding of AI scribe adoption by demonstrating how behavioral drivers across capability, opportunity, and motivation shape implementation. Practical barriers at the system level, particularly time and capacity constraints, were prominent, while enablers included high-quality training, peer and leadership support, clear communication, and access to appropriate resources. Individual motivational factors, such as perceptions of the AI scribe’s value, safety, reliability, and openness to change, as well as attitudes toward AI and digital solutions, also influenced uptake. Building on these insights, the paper discusses targeted interventions to help health services embed AI scribes into allied health practice at both the system and individual levels, thereby unlocking the potential of this promising technology.