Abstract
Objectives
Allied Health Professionals (AHPs) represent the second largest healthcare workforce in the United Kingdom, yet their perspectives on artificial intelligence (AI) remain underrepresented in the literature. As national policy commits to making the NHS “the most AI-enabled care system in the world”, understanding AHP engagement with AI is essential for effective workforce development and implementation.
Methods
An anonymous cross-sectional survey was conducted among AHPs attending a 2025 regional Research and Innovation Conference in the United Kingdom. The survey explored demographics, AI familiarity and use, perceived benefits and concerns, and views on regulation. Data were analysed using descriptive statistics, Kruskal–Wallis and chi-square tests, and exploratory binary logistic regression, following STROBE and CHERRIES guidelines.
Results
Among 162 respondents from nine AHP professions, 54.9% reported personal use of large language models, while only 21.7% used AI professionally, mainly for administrative tasks. Healthcare-specific AI familiarity was low (median = 2/5). In exploratory regression analysis familiarity with healthcare AI emerged as the only significant predictor of comfort with AI-supported clinical decision-making (OR = 2.28, 95% CI [1.49–3.47], p < .001); age, personal AI use, and discipline were not independent predictors. Radiographers demonstrated greater healthcare AI familiarity than dietitians (p < .001) and occupational therapists (p = .021). Most participants viewed AI as an assistive tool requiring human oversight (93.2% wanted override ability), and fewer than 20% were comfortable with AI making patient-care decisions. Access to education and training was identified as a key enabler by 90.1% of respondents.
Conclusions
Workplace integration of AI remains limited despite increasing accessibility of AI technologies. Comfort with clinical AI use was associated with familiarity with healthcare-specific applications and awareness of their potential uses. These findings suggest implementation efforts may benefit from improving foundational understanding of AI and its applications, alongside clear accountability frameworks.
Keywords
Introduction
In the United Kingdom (UK), Allied Health Professionals (AHPs) form the second largest professional workforce in health and social care, delivering vital multidisciplinary services across primary, secondary, and community settings. 1 Fourteen professional groups are broadly recognised as AHPs in the United Kingdom. 2 These professional groups are facing mounting pressure related to increasing workloads, clinical complexity, workforce shortages, and increasing expectations around digital capability. Against this backdrop, artificial intelligence (AI) has emerged as a prominent feature of national health policies, with the UK government pledging to make the NHS “the most AI-enabled care system in the world”. 3 Proponents suggest AI has the potential to address longstanding healthcare challenges by improving diagnostics, supporting personalisation of treatment, and reducing waiting times. However, the rapid pace of AI implementation raises critical questions about workforce readiness and engagement. Previous digital health initiatives have demonstrated that technology adoption often fails when systems are implemented without adequate engagement from those expected to use them.4,5 Understanding if and how AHPs use AI, their concerns, and their support needs is vital to ensure AI enhances healthcare practice and service delivery rather than creating additional challenges.
Research exploring health professional attitudes towards AI suggests a cautious optimism alongside significant knowledge gaps and professional concerns. A recent systematic review of 72 studies involving over 15,000 health professionals demonstrated that while AI’s potential to improve efficiency, diagnostic accuracy, and workload management is recognised, significant barriers exist across individual, interpersonal, institutional, and policy levels. 6 At an individual level, limited AI knowledge, inadequate education and training, and fears of job replacement or deskilling emerged as consistent barriers. At institutional levels, concerns focused on potential clinical errors, lack of organisational readiness, and unclear liability frameworks. Across the studies health professionals expressed significant concerns about AI’s inability to account for patient complexity, the potential loss of therapeutic relationships, and data security issues. Notably, this evidence base is derived almost exclusively from studies involving medical specialists, particularly in radiology (20 studies). While the review included some studies involving physical therapists, radiographers and pharmacists, AHP perspectives remain substantially underrepresented, limiting our understanding of how this diverse workforce perceives and engages with AI. The Health Foundation’s 2024 survey of NHS staff, which included an AHP subgroup, reported broad workforce support for AI in patient care, with familiarity positively associated with attitudes towards AI, however profession-specific data for individual AHP groups was not reported. 7
Warrington and Holm 8 surveyed 211 UK doctors and found that whilst 55.7% reported using AI, only 7.1% had received formal training. Doctors overwhelmingly called for regulatory frameworks (98% agreement) and professional guidance from royal colleges (95% agreement). Notably, 37% of AI-using doctors reported using it to write reflective portfolio pieces raising concerns about implications for professional learning. 9 Similarly, Hoffman et al. 10 surveyed 231 Australian AHPs and found that 80.1% had never used AI and 87% possessed little to no knowledge about it, 77.1% reported workforce knowledge and skills as a key implementation barrier. Significant interprofessional variation emerged, with pharmacists substantially more likely than occupational therapists, physiotherapists, or social workers to anticipate AI affecting their roles (p=0.009).
However, both studies were conducted during the early stage of Large Language Models (LLMs) like ChatGPT and Claude becoming mainstream. Since then, AI accessibility has changed rapidly. ChatGPT launched in November 2022, and usage among Americans roughly doubled between summer 2023 and 2024. 11 Workplace integration tools like Microsoft Copilot became more widely available in the UK in November 2023. This rapid evolution may have altered health professional perspectives considerably as AI has transitioned from novel to somewhat routine in day-to-day life and workplaces. Given that most AHPs have not yet used AI professionally, this study takes an exploratory approach to capture current attitudes to help inform AI research and development relating to AHP clinical practice and education. Moreover, UK profession-specific data characterising attitudes across the broader AHP workforce remains limited. 12
This study addresses these evidence gaps by surveying AHPs attending a regional UK AHP conference.
The study aimed to (i) describe AHP awareness, knowledge, and current use of AI, (ii) examine attitudes toward AI in healthcare, including perceived relevance, enablers, and concerns and (iii) explore factors that may be associated with comfort using AI in clinical practice.
Methods
Study design
An anonymous cross-sectional digital survey of AHPs was conducted in September 2025 during a professional conference held in Northern Ireland. The study followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting observational studies and the (Checklist for Reporting Results of Internet E-Surveys) guidelines for internet-based survey research. 13
Ethical approval
Ethical approval was obtained from Ulster University Nursing and Health Research Ethics Filter Committee (FCNUR-25-061). The survey opened with a Participant Information Sheet (PIS) detailing the study purpose, voluntary participation, and data handling procedures. Participants provided electronic informed consent via mandatory checkboxes before accessing survey questions.
Participants and sampling
The target population was the fourteen professional groups recognised as AHPs: Physiotherapists, Occupational therapists, Speech and language therapists, Dietitians, Art therapists/art psychotherapists, Prosthetists, Orthotists, Music therapists, Orthoptists, Podiatrists, Diagnostic radiographers, Therapeutic radiographers, Paramedics and Drama therapists. No formal sample size calculation was conducted; this exploratory study aimed to capture the range of AHP perspectives and generate hypotheses for future investigation rather than test specific a priori hypotheses. A convenience sample was recruited through the 2025 AHP conference. Nine AHP professions were represented in the final sample (see results). All AHPs were eligible to participate, three respondents who were not AHPs were excluded from the analysis.
Recruitment and survey access
Information about the survey was included in pre-conference advertising materials. During the conference (September 2025), a QR code linking to the survey was displayed on holding slides between presentations. To allow participation from AHPs who could not attend on the day, the survey remained open for two weeks post-conference, and the QR code was re-circulated through the same communication channels used for the initial promotion.
Survey development and piloting
The survey questionnaire was developed following a literature review that identified key themes in health professional attitudes toward AI including concerns about safety, autonomy, and data security. Selected questions were adapted from those used in previous surveys of health professionals.8,10,12 The draft survey was piloted with four AHPs from two disciplines (physiotherapy n=2, radiography n=2); feedback was provided on clarity, survey length and structure. Revisions included rewording ambiguous questions, removal of repetitive questions and mandatory completion of key questions. The final questionnaire was divided into six sections exploring (i) demographics, (ii) challenges in health services, (iii) familiarity with AI (self-rated knowledge and awareness of AI applications), (iv) views on AI in current role (including comfort with AI integration and perceived relevance to practice), (v) barriers and enablers to use of AI (vi) AI ethics and concerns. The survey included a mixture of Likert scales, multiple-select questions and free text responses.
Survey administration
The survey was administered via Jisc Online Surveys, a secure, GDPR-compliant platform widely used in UK higher education institutions. The survey was designed to be completed in approximately 10 minutes. Unique response IDs, timestamps and responses were screened for any obvious duplication. Adaptive questioning (skip logic) was used to improve survey efficiency, ensuring participants only answered questions relevant to them. Participants could return to previous questions. To protect anonymity, no personally identifiable information was collected, IP addresses were not tracked, and no login credentials were required. No incentives were offered for participation.
Response rate and completeness
A total of 165 individuals accessed the survey, confirmed they had read a participant information sheet and agreed to take part. Three responses from non-AHPs were excluded from the analysis resulting in a final sample of 162. We did not identify any obvious repeated entries. All submitted responses met the completion threshold of (≥50% of questions answered). A completed CHERRIES checklist is included [Supplementary File 1]. As the survey was accessible across different channels the total number of eligible individuals receiving the link cannot be determined therefore no formal response rate was calculated.
Data analysis
Quantitative data
Survey data were exported from Jisc Online Surveys and analysed using IBM, SPSS Version 30.0.0 software. All participants who responded to a given question were included in the analysis for that question, resulting in varying sample sizes across items due to skip logic and non-response options. Missing data were handled through complete case analysis for each item.
Descriptive statistics: frequencies and percentages were calculated for categorical variables. For ordinal Likert-scale data medians and interquartile ranges (IQR) were reported given the non-normal distribution of these scales.
Subgroup comparisons: For subgroup analyses, age was collapsed into three categories (under 35, 35–54, and 55+ years) to ensure adequate cell sizes. Professional groups were limited to the five largest disciplines (physiotherapists, dietitians, radiographers, occupational therapists, and speech and language therapists, n=143), with professions having fewer than 10 respondents excluded due to insufficient sample size.
Differences between professional groups were examined using Kruskal-Wallis tests for ordinal variables and chi-square tests for categorical variables. Where Kruskal-Wallis tests indicated significant differences (p<.05), Dunn’s post-hoc pairwise comparisons with Bonferroni correction were conducted to identify specific group differences. Where chi-square test assumptions were violated (>20% of cells with expected counts <5), Fisher’s exact test was used to confirm results.
Binary logistic regression was conducted as an exploratory, hypothesis generating analysis to identify potential predictors of comfort with using AI to support clinical decision-making among the five largest professional groups (n=143). All predictor variables - healthcare-specific AI familiarity, personal use of large language models, age category, and professional discipline were entered simultaneously into the model. Multicollinearity was assessed through examination of bivariate correlations between predictors. Overall model fit was assessed using the Omnibus test and Nagelkerke R2 statistic, Hosmer-Lemeshow goodness of fit test and classification accuracy.
Qualitative data
Free text responses to open ended questions were analysed using inductive content analysis. 14 One researcher read all responses, identified recurrent themes and patterns and categorised responses accordingly. Representative quotes were selected to illustrate key themes.
Results
Sample characteristics
Participant characteristics (N=162).
*Includes diagnostic and therapeutic radiographers.
†Includes art/music/drama therapist and orthoptist.
‡Multiple responses possible.
Self-reported AI knowledge, familiarity and current use
Self-reported AI knowledge, use, and training needs (N=162).
*Rated on 5-point scale (1=lowest, 5=highest).
†Participants selected up to three options.
Over half of participants (54.9%, n=89) reported personally using LLMs such as ChatGPT outside of work contexts, use of AI within professional roles was much lower with only 21.7% (n=35) reporting use of AI or machine-assisted decision-making in their work and 73.9% (n=119) indicating no current AI use. Among those using AI professionally (n=35), free-text responses revealed predominant use in administrative tasks including report writing and documentation. A smaller number of radiographers specifically mentioned AI-aided diagnostic tools for image analysis and predictive modelling. Participants reported using AI for research, for example Copilot and to support their understanding of data analysis.
Significant professional variation emerged in healthcare-specific AI familiarity (Kruskal-Wallis H=21.203, df=4, p<.001, n=144). Radiographers reported the highest familiarity (mean rank=97.23), followed by speech and language therapists (76.31), physiotherapists (75.71), occupational therapists (60.78), and dietitians (50.45). Post-hoc pairwise comparisons with Bonferroni correction revealed that radiographers had significantly higher healthcare AI familiarity than both dietitians (adjusted p<.001) and occupational therapists (adjusted p=.021). No other pairwise comparisons reached statistical significance.
Personal use of large language models also differed significantly by age group (χ2=10.802, df=2, p=.005, n=161). Usage rates were highest among those under 35 years (79.3%), compared with (51.8%) among those aged 35–54 years and (33.3%) among those aged 55 years and older.
Attitudes toward AI in healthcare
Participants demonstrated cautiously supportive attitudes toward AI, with strong emphasis on maintaining human oversight (Figure 1). Near-universal agreement was observed for statements emphasising clinician autonomy: 93.2% (n=151) wanted to retain ability to override AI suggestions, 92.0% (n=149) wanted to understand how AI reaches recommendations, and 88.3% (n=143) would trust AI more if they remained responsible for final decisions. Allied health professional attitudes towards AI use in healthcare (N=162).
Support for AI as a supportive tool was evident: 73.1% (n=117) agreed AI could be useful for monitoring treatment progress, 69.1% (n=112) thought it could identify things clinicians might miss, and 66.5% (n=107) believed it could help guide treatment planning. However, comfort with AI in decision-making showed important distinctions. While 45.3% (n=73) were comfortable using AI to support their clinical decision-making, only 19.9% (n=32) would be comfortable with AI making autonomous decisions about patient care, with 49.7% (n=80) disagreeing. This reflects acceptance of AI as an assistive tool while rejecting autonomous AI decision-making. I think clinicians always need to maintain autonomy and are ultimately responsible for decisions
Additional benefits identified by participants included improved time efficiency for specific tasks, support for innovative workflows and to enhanced clarity of written communication.
A majority 58.6% (n=95) expressed concerns that AI could negatively affect the clinician-patient relationship.
Predictors of clinician comfort in using AI to support clinical decision-making
Binary logistic regression was conducted among the five largest professional groups (n=143) to identify predictors of comfort in using AI to support clinical decision-making. All predictor variables (healthcare-specific AI familiarity, personal use of LLMs, age category, and professional discipline) were entered simultaneously into the model. Bivariate correlations between predictor variables were examined, all correlations were weak (r<0.3), indicating multicollinearity was not a concern.
The overall model was statistically significant (χ2=28.947, df=8, p<.001) and explained approximately 24.7% of the variance in comfort with AI (Nagelkerke R2=.247) and correctly classified 70.4% of cases. The Hosmer-Lemeshow test indicated good model fit (χ2=10.736, df=8, p=.217).
Healthcare-specific AI familiarity emerged as the only significant independent predictor (OR=2.28, 95% CI [1.49–3.47], p<.001). For each one-unit increase in healthcare AI familiarity, AHPs were more than twice as likely to report comfort in using AI to support their clinical decisions. Personal use of LLMs (p=.173), age (p=.750), and professional discipline (p=.773) did not independently predict comfort with using AI when controlling for healthcare AI familiarity.
Perceived relevance and anticipated impact
Current perceived relevance of AI to professional roles showed mixed views (median=3, IQR=3–4 on a 5-point scale), with just under half (46.9%) rating AI as quite to very relevant to their current role. However, anticipated impact over the next five years was considerably higher (median=4, IQR=3–5), with 71.0% expecting significant to major impact on their field.
Potential applications and enablers for integration
When asked to select up to three areas where AI could be most useful, participants most frequently identified education and training (67.3%, n=109) and administrative work (64.8%, n=105). Clinical applications commonly selected included diagnostic assistance (45.7%, n=74), triage and prioritisation (42.6%, n=69), and monitoring rehabilitation progress (32.1%, n=52). Notably, only 1.2% (n=2) indicated they did not think AI should be used in healthcare at all. Anything that helps improve effectiveness and efficiency of time and reduces human error is helpful
Participants overwhelmingly prioritised access to training and education as essential for AI integration (90.1%, n=146), followed by clear evidence of improved outcomes or efficiency (74.7%, n=121) and easy-to-use tools (63.6%, n=103). Professional body endorsement was selected by 45.1% (n=73). Interest in receiving AI training was high, with 78.8% expressing very to extremely high interest. Only 3.1% (n=5) indicated they were unlikely to use AI regardless of enabling factors. We get little training on digital skills but use technology all the time. More training would be of benefit to many
Concerns and accountability
Open-ended responses about AI concerns (n=113 substantive responses) revealed several themes. The most frequently expressed concerns related to accountability and liability for AI-supported decisions causing harm. Data security, confidentiality, and GDPR compliance were raised repeatedly with many identifying uncertainties around policy and usage permissions for AI in the workplace and the potential for ethical issues if AI are not properly governed. Many worried about potential loss of clinical reasoning skills if clinicians became over-reliant on AI. Impact on the clinician-patient relationship was a common theme, with concerns about loss of empathy and personalised care. Some participants worried about job security and workforce deskilling. Several emphasised that AI should never replace human oversight and that final decision-making must remain with clinicians. I am concerned clinicians will become solely reliant on AI and lose clinical reasoning skills and become complacent”
Regarding accountability for AI-related harm, the largest proportion (45.9%, n=73) favoured shared responsibility between all parties involved, with smaller proportions attributing primary responsibility to the healthcare organisation (13.2%, n=21), the clinician (12.6%, n=20), or the AI developer (10.1%, n=16). Nearly one-fifth (18.2%, n=29) were unsure.
When asked where AI should never be used (n=98 responses), participants identified end-of-life decisions, final diagnostic determinations, communication of serious diagnoses, and complex ethical situations. The rationale focused on AI’s inability to grasp nuances in clinical reasoning and moral judgment. Participants emphasised that AI must remain a supportive tool, never replacing human decision-making in situations requiring clinical judgment, empathy, or consideration of individual circumstances. Documentation of integration of AI is essential in patient treatment/management so that decision-making rationale is traceable e.g. in the event of a misdiagnosis the AI element should be distinguishable from the human element. I feel that clinical experience cannot be replaced and AI may not always pick up on personal or nuanced variables that are relevant”
Discussion
Principal findings
This study examined attitudes, familiarity, and professional use of AI among AHPs attending a regional professional conference in the UK, offering multidisciplinary insights at a time of proposed rapid AI expansion within healthcare. Among 162 respondents representing nine professions, personal engagement with LLMs was common (54.9%), but only 21.7% reported using AI within their professional role, most frequently for administrative or educational purposes. Healthcare-specific AI familiarity emerged as the only independent predictor of comfort in using AI to support clinical decision-making in exploratory regression analysis (OR = 2.28, 95% CI [1.49–3.47], p < .001), with demographic factors and general AI use showing no significant association. These results may indicate that contextual understanding of how AI operates within one’s own scope of practice may be associated with confidence in using AI in the workplace.
Patterns of AI use and professional variation
Although personal use of AI has become more widespread possibly driven by a push from established platforms and societal curiosity, integration into clinical practice remains limited. Respondents described workplace use of AI mainly for ‘low risk’ tasks such as report writing, communication, or data organisation. This pattern is similar to other surveys that suggest clinicians engage with AI for administrative efficiency but show greater caution when it informs diagnosis or treatment.8,10 This caution appears to reflect uncertainty about accuracy, accountability, and the implications for professional autonomy. Radiographers in this study, who demonstrated significantly greater healthcare-specific AI familiarity than occupational therapists (p = .021) and dietitians (p < .001), provided one of the few examples of AI use in direct clinical work, typically through imaging systems already subject to regulatory evaluation. This finding aligns with radiology’s position as an early adopter of AI, with imaging accounting for roughly three-quarters of all FDA approved AI-enabled medical devices15,16 and radiographers reporting early exposure and generally positive attitudes. 12 Across other AHP disciplines, comparable regulation and clinical evidence remains scarce.
In this study, 21.7% of AHPs reported using AI professionally, while 66% reported minimal awareness of healthcare-specific AI applications. Hoffman et al. 10 reported similar professional AI use (19.9%) but higher rates of minimal awareness (87%). While these differences may reflect the later timing of our data collection, after generative AI became more widely accessible in workplaces, they may also reflect differences in healthcare context, professional mix, or sampling approach. The American Medical Association 16 reported a rise in physician AI use from 38% in 2023 to 66% in 2024, while UK surveys found lower rates, with around 20% of GPs 17 and 29% of doctors across specialties 18 reporting workplace AI use. While direct comparison is limited by differences in methodology and populations, these studies indicate that clinical AI integration varies considerably across healthcare professions and settings.
The broadly supportive attitudes observed in this study are consistent with NHS workforce-level data indicating mostly positive attitudes towards AI within healthcare. 7 However, beyond radiographers, 12 published UK data disaggregating AI attitudes and familiarity by AHP professions are limited, limiting broader comparisons. However, disaggregating findings by profession within the present study revealed interprofessional variation in healthcare-specific AI familiarity which may have implications for profession-specific approaches to AI integration.
Evidence, regulation, and professional caution
Caution toward AI-supported decision-making likely reflects overlapping influences at individual, professional, organisational, and system levels. In line with our findings, qualitative studies consistently report clinician concerns about accountability, explainability, and erosion of professional judgement.6,19,20 Participants in this study emphasised the need for evidence of effectiveness before adoption, reinforcing that availability of technology is insufficient to ensure implementation particularly when it is determining patient intervention protocols/pathways. The British Medical Association recently reinforced this position, supporting doctors who refuse to use untested or non-evidence-based AI systems. 21 With the exception of radiography, few AHP specific AI tools have been rigorously evaluated or subjected to robust regulatory scrutiny this may reflect the impact and influence of radiologists who are medical professionals but closely aligned to radiographers, who are AHPs.
There is potential tension between policy ambition and the readiness of health professionals for AI integration. The UK Government’s Fit for the Future: 10-Year Health Plan for England 3 pledges to make the NHS “the most AI-enabled care system in the world.” While this suggests a commitment to digital transformation, it also highlights the need for robust evidence, governance and workforce capability to ensure safe and effective implementation. Frameworks such as the Human Organisation Technology (HOT) model and NHS Digital Transformation guidance stress these features are critical for implementation success. 22 However, several reviews highlight that AI tools are often introduced ahead of robust validation.4,6 Participants in the present study expressed reluctance to use AI without clear regulatory approval or evidence of benefit, reflecting broader concerns about technology implementation outpacing evaluation.
Education, training, and workforce readiness
Strengthening confidence in AI may require education that enables clinicians to interpret evidence and apply technology safely. In this study, 90% of participants identified education and training as essential for AI integration, reflecting a consistent pattern across the wider healthcare literature, where lack of self-reported AI knowledge and training emerge as barriers to adoption.6,8,10 Evidence from systematic reviews on AI education suggests that studies are focused on medical and nursing education with much less evidence on AHP education.23,24 Integrating AI education at all levels of AHP education and as part of continuing professional development frameworks could improve awareness of utility, efficacy, bias, ethical considerations and critical appraisal. These competencies are essential for safe and accountable use of AI in clinical practice and align with guidance published by the Health and Care Professions Council (HCPC), the regulatory body for AHPs in the UK, which has identified AI literacy, academic integrity, and integration of emerging technology into curricula as priority areas for education providers. 25
Strengths and limitations
This study provides contemporary, multidisciplinary insight into AHP attitudes toward AI during a period of rapid technological change in healthcare. The inclusion of nine professional groups enabled interprofessional comparisons revealing interprofessional comparisons, revealing variation in healthcare specific AI familiarity which may have implications for how education and implementation strategies may need to be tailored across the AHP workforce. The combination of quantitative and qualitative data supported both descriptive characterisation of the attitudinal landscape and exploratory identification of factors associated with comfort in the clinical use of AI. However, several limitations warrant consideration. Convenience sampling through a professional conference may have favoured participants with greater interest in innovation or digital technology. The gender distribution (84.5% female) is broadly consistent with UK AHP workforce demographics where women comprise approximately 80% of the allied health workforce. 26 However, the predominance of more experienced clinicians (>10 years) may limit generalisability to more recently qualified practitioners whose attitudes and training experiences may differ. The cross-sectional design limits causal inference. Self-reported measures may not accurately capture competence, and smaller subgroup sizes restrict detailed comparison. Constructs were measured using single items adapted from prior surveys rather than validated multi-item scales, limiting measurement precision and reliability metrics such as Cronbach’s alpha are not applicable to single-item measures. Potential confounders including years of experience and highest qualification were not included in the regression model, and no sensitivity analyses were conducted. No formal sample size calculation was conducted as the total eligible population could not be established limiting precision of effect estimates and precluding assessment of potential non-response bias. Finally, given the pace of AI development, results represent a specific snapshot that may evolve rapidly.
Conclusions and future directions
AHPs in this survey displayed cautious optimism toward AI. Personal use of generative tools was more common in participants but clinical integration remained limited. Comfort with use appears to be associated with self-reported healthcare-specific knowledge and the availability of evidence demonstrating safety and effectiveness. Radiographers’ greater familiarity likely reflects the concentration of validated AI applications within imaging, highlighting a potentially uneven exposure across AHPs. These findings highlight the potential value of evidence generation, professional education, and governance frameworks to support AHP engagement with AI. Future studies should examine the longitudinal impact of AI training and regulatory guidance, explore patient perspectives on AHP-led AI use, and monitor how attitudes shift as validated tools become more accessible. Ensuring that AHPs are equipped to interpret and apply AI evidence responsibly will be important for achieving safe, effective, and ethically grounded AI use in practice.
Footnotes
Acknowledgement
The authors would like to acknowledge Professor Chris Bleakley, Ulster University for advice regarding statistical analysis.
Ethical considerations
Ethical approval was obtained from Ulster University Nursing and Health Research Ethics Filter Committee (FCNUR-25-061).
Consent to participate
Participants consented to data being used for this study and resulting publications but were not consented for public data sharing.
Author contributions
All authors contributed to the conceptualisation of the study, designing the methodology and plan of investigation, all authors reviewed and edited the final manuscript.
Conceptualisation (All)
Formal analysis JM
Investigation (All)
Methodology (All)
Project administration (JM)
Resources (All)
Validation (JM, KP, SMcF)
Visualization (All)
Writing – original draft (JM, KP, SMcF)
Writing – review & editing (All).
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The survey instrument and statistical syntax are available from the corresponding author on reasonable request.
Guarantor
Dr Joanne Marley accepts responsibility for the overall integrity of the work, had access to the data, ensured all authors meet authorship criteria, confirms ethical approvals and disclosures, and controlled the decision to publish.
Declaration/Data sharing
Editorial assistance from ChatGPT (OpenAI, GPT-5, 2025) was used for minor grammar refinement of sections of the manuscript and reducing word count. The authors reviewed and verified all content to ensure accuracy and integrity. The AI tool was not used for data analysis, interpretation, or generation of original ideas.