Paper deep dive
Multi-Level Barriers to Generative AI Adoption Across Disciplines and Professional Roles in Higher Education
Jianhua Yang, Kerem Ăge, Adrian von MĂźhlenen, Abdullah Bilal Akbulut, Tanya Suzanne Carey, Chidi Okorro
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 92%
Last extracted: 3/31/2026, 1:59:27 AM
Summary
This study investigates the structural barriers to Generative AI (GenAI) adoption in higher education by surveying 272 staff members at a Russell Group university. It moves beyond individual-level factors (like perceived usefulness) to analyze how disciplinary contexts (STEM vs. non-STEM) and institutional roles (academic vs. professional services) shape adoption. The findings reveal that non-STEM academics primarily face ethical and cultural barriers, while STEM and professional services staff are more constrained by institutional, governance, and infrastructure issues.
Entities (5)
Relation Signals (3)
Non-STEM Academics â reportsbarrier â Ethical and Cultural Barriers
confidence 90% ¡ non-STEM academics primarily report ethical and cultural barriers related to academic integrity
STEM and PSs Staff â reportsbarrier â Institutional and Infrastructure Constraints
confidence 90% ¡ STEM and PSs staff disproportionately emphasize institutional, governance, and infrastructure constraints
Disciplinary Context â shapes â GenAI Adoption
confidence 85% ¡ we examine how disciplinary contexts and institutional roles shape perceived barriers
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Generative Artificial Intelligence (GenAI) is rapidly reshaping higher education, yet barriers to its adoption across different disciplines and institutional roles remain underexplored. Existing literature frequently attributes adoption barriers to individual-level factors such as perceived usefulness and ease of use. This study instead investigates whether such barriers are structurally produced. Drawing on a multi-method survey analysis of 272 academic and professional services (PSs) staff at a Russell Group university, we examine how disciplinary contexts and institutional roles shape perceived barriers. By integrating multinomial logistic regression (MLR), structural equation modelling (SEM), and semantic clustering of open-ended responses, we move beyond descriptive accounts to provide a multi-level explanation of GenAI adoption. Our findings reveal clear, systematic differences: non-STEM academics primarily report ethical and cultural barriers related to academic integrity, whereas STEM and PSs staff disproportionately emphasize institutional, governance, and infrastructure constraints. We conclude that GenAI adoption barriers are deeply embedded in organizational ecosystems and epistemic norms, suggesting that universities must move beyond generalized training to develop role-specific governance and support frameworks.
Tags
Links
- Source: https://arxiv.org/abs/2603.27052v1
- Canonical: https://arxiv.org/abs/2603.27052v1
Trouble viewing inline? Open PDF directly â
Full Text
60,700 characters extracted from source content.
Expand or collapse full text
1 Multi-Level Barriers to Generative AI Adoption Across Disciplines and Professional Roles in Higher Education Jianhua Yang 1* , Kerem Ăge 2 , Adrian von MĂźhlenen 3 , Abdullah Bilal Akbulut 4 , Tanya Suzanne Carey 1 , Chidi Okorro 1 1 WMG, the University of Warwick, UK; 2 Department of Politics and International Studies, the University of Warwick, UK; 3 Department of Psychology, the University of Warwick, UK; 4 Birmingham Business School, the University of Birmingham, UK * Correspondence: Jianhua.Yang@warwick.ac.uk Abstract Generative Artificial Intelligence (GenAI) is rapidly reshaping higher education, yet barriers to its adoption across different disciplines and institutional roles remain underexplored. Existing literature frequently attributes adoption barriers to individual-level factors such as perceived usefulness and ease of use. This study instead investigates whether such barriers are structurally produced. Drawing on a multi-method survey analysis of 272 academic and professional services (PSs) staff at a Russell Group university, we examine how disciplinary contexts and institutional roles shape perceived barriers. By integrating multinomial logistic regression (MLR), structural equation modelling (SEM), and semantic clustering of open-ended responses, we move beyond descriptive accounts to provide a multi-level explanation of GenAI adoption. Our findings reveal clear, systematic differences: non-STEM academics primarily report ethical and cultural barriers related to academic integrity, whereas STEM and PSs staff disproportionately emphasize institutional, governance, and infrastructure constraints. We conclude that GenAI adoption barriers are deeply embedded in organizational ecosystems and epistemic norms, suggesting that universities must move beyond generalized training to develop role-specific governance and support frameworks. Keywords Generative AI; higher education; technology adoption; professional services; structural equation modelling 2 Introduction Generative artificial intelligence (GenAI) is rapidly reshaping higher education (HE) (OâDea, 2024). Large language models (LLMs) are now used in teaching preparation, assessment design, student support, administration, and institutional communication. They promise efficiency and pedagogical innovation (Daniel et al., 2025), yet they also raise concerns about academic integrity, bias, privacy, workload, and the erosion of core academic skills (Abbas et al., 2024; Sullivan et al., 2023). Adoption is therefore uneven and often contested across universities. Many studies explain technology adoption primarily through individual-level factors such as perceived usefulness, ease of use, literacy, or attitudes (Ghimire & Edwards, 2024; ThĂźs et al., 2024; Williams et al., 2015). While valuable, these approaches understate the institutional complexity of HE. Universities are structured by disciplinary norms, professional identities, governance arrangements, and regulatory constraints. Decisions about GenAI use are therefore not simply personal choices; they are embedded within institutional roles and subject cultures. This study asks whether barriers to GenAI adoption in HE are primarily individual-level issues, or whether they are structurally produced by institutional roles, disciplines, norms, and ethical expectations. This distinction has direct policy implications. If barriers are individual, training interventions may suffice. If they are structural, governance design and role-specific support become central. To address this question, we draw on a 2025 survey of staff at a Russell Group university on GenAI use and barriers, completed by 272 respondents. The survey covers both academic and professional services (PSs) roles and captures literacy, attitudes, job threat perceptions, institutional guidance and support, ethical concerns, and perceived barriers across teaching and workplace contexts. Methodologically, we adopt a multi-method approach to survey analysis. We combine descriptive statistics, multinomial logistic regression, structural equation modelling (SEM), and clustering of free-text responses to analyse both systematic differences and underlying mechanisms. This design allows us to connect reported barriers to disciplines and professional roles. Our findings show clear and systematic differences in reported barriers across disciplines and roles. STEM (science, technology, engineering, and mathematics) staff are more likely to report institutional and individual barriers, whereas non-STEM staff report higher ethical and cultural barriers in absolute terms. PSs staff differ significantly from academics, particularly in relation to institutional constraints they face. We also find that individual factors such as literacy, attitudes, and job threat matter, but they operate within structured disciplinary and organizational contexts. These results demonstrate that GenAI adoption barriers reflect academic disciplinary cultures, professional roles, and ethical expectations as much as individual skill or perception. 3 Our contribution is threefold. First, we provide one of the few systematic cross-disciplinary analyses of GenAI barriers, directly comparing STEM and non-STEM contexts within the same institution. Second, and more distinctively, we extend the analysis beyond teaching academics to include PSs staff. This group plays a critical role in university operations yet is largely absent from the GenAI adoption literature. To our knowledge, no prior study offers a large-scale comparative analysis of both academic and PSs roles within a single institutional setting. Third, by integrating quantitative modelling with qualitative clustering, we move beyond descriptive accounts of barriers and develop a multi-level explanation of GenAI adoption. The remainder of the paper proceeds as follows: the next section reviews relevant theoretical and empirical literature, followed by a description of the research design and methods; we then present the results and a discussion of our findings. Barriers to GenAI adoption in HE GenAI has generated both optimism and unease in HE. Advocates argue that it may significantly reshape teaching and learning practices, enhancing feedback, personalization, and efficiency (Khan, 2024). Critics, however, emphasize risks to academic integrity, authorship, and professional standards (Bearman et al., 2024). Some scholars contend that many of these risks stem not from the technology itself but from assessment systems designed for a pre-GenAI environment, and therefore call for redesign rather than prohibition (Ardito, 2025; Evangelista, 2025). Others warn that habitual reliance on GenAI may weaken the development of critical thinking and writing skills (Lindebaum & Fleming, 2024). While much of this debate is framed normatively, empirical research increasingly points to a heterogeneous set of barriers shaping adoption in practice (Batista et al., 2024). Synthesizing this literature suggests that these barriers operate across overlapping individual, ethical, social, and institutional dimensions. We begin with individual barriers. Classic individual-level technology adoption frameworks such as the Technology Acceptance Model (TAM) emphasize perceived usefulness and ease of use as key determinants of uptake (Davis, 1989; Venkatesh et al., 2003; Williams et al., 2015). These constructs remain highly relevant in GenAI contexts (Ghimire & Edwards, 2024; ThĂźs et al., 2024), particularly in explaining variations in literacy, confidence, and perceptions of effectiveness (Al-Abdullatif, 2024). Empirical studies identify a range of individual-level constraints, including limited AI literacy among staff and students (Mah & GroĂ, 2024), insufficient training and low self-efficacy, time pressures that restrict experimentation (Watermeyer et al., 2025), and concerns about inaccuracy or misinformation (Shata, 2025). Yet emerging scholarship cautions against reducing barriers to individual cognition alone. Models centred narrowly on attitudes risk overlooking the 4 organizational, cultural, and societal conditions that shape how such perceptions are formed and acted upon (Ayanwale, 2024; FakhrHosseini et al., 2024; Shrivastava, 2025). Second, a substantial body of research demonstrates that resistance to AI is frequently grounded in ethical and socio-cultural concerns rather than purely technical limitations (Al-Shabandar et al., 2024; Dabis & CsĂĄki, 2024; Essien et al., 2024; Khlaif et al., 2024; Lan et al., 2025), dimensions that are only partially captured within Unified Theory of Acceptance and Use of Technology (UTAUT) style models (Nikolic et al., 2024). Documented concerns include privacy and data security (Chan & Hu, 2023; Gasaymeh et al., 2024), academic integrity and plagiarism (Sullivan et al., 2023), intellectual property rights (Asad & Ajaz, 2024), pedagogical orientation (Cabero-Almenara et al., 2024), skill erosion and overreliance (Chiu, 2024), and threats to intellectual authenticity and creativity (Shata, 2025). Scholars also highlight anxieties about dehumanization and the erosion of professional judgement, distrust in AI outputs, ethical ambivalence regarding training data and monetization (Watermeyer et al., 2025), inequitable access (Jin et al., 2025; Valdivieso & GonzĂĄlez, 2025), and cross-cultural variation in acceptance (Wang et al., 2023; Yusuf et al., 2024). Taken together, these concerns indicate that GenAI unsettles normative understandings of academic labour and professional responsibility. Adoption decisions are therefore shaped not only by instrumental evaluations of utility, but also by ethical expectations, socio-cultural contexts, disciplinary norms, and professional identities. Beyond these barriers, institutional structures and norms play a decisive role in shaping adoption conditions. A growing body of research identifies structural barriers such as the absence of clear institutional guidance (An et al., 2025; Michel-Villarreal et al., 2023; Nguyen, 2025), limited development of inclusive governance frameworks (Valdivieso & GonzĂĄlez, 2025), and cost and infrastructure constraints, including bandwidth limitations and digital divides (Watermeyer et al., 2025). Studies also point to gendered differences in access (Al-Samarraie et al., 2025; Strzelecki & ElArabawy, 2024), compliance pressures linked to data protection regimes such as GDPR (General Data Protection Regulation) and the EU AI Act, and the need for formal oversight and ethics governance mechanisms (Jayaram et al., 2024). Even where high level policy encouragement exists (ChristâBrendemĂźhl, 2025), translating broad guidance into local, department level practice remains challenging (CoffĂŠ et al., 2026; Lee et al., 2025). Taken together, these findings suggest that GenAI adoption is embedded within governance environments and institutional capacities. Perceived barriers may therefore reflect structural constraints and organizational design rather than simple individual reluctance. Recent scholarship therefore argues that adoption in complex educational settings should be analysed through multi-level frameworks that integrate individual, ethical, and institutional dimensions rather than isolating beliefs 5 or attitudes (Ayanwale, 2024; FakhrHosseini et al., 2024; Shrivastava, 2025). Yet two important gaps remain. First, cross disciplinary comparisons are still relatively rare (Zhao et al., 2025). Second, the PSs workforce is largely absent from empirical analyses, despite its central role in governance, compliance, infrastructure, and operational workflows (Kutty et al., 2024). This omission constrains our ability to distinguish clearly between barriers rooted in individual cognition and those produced by structural or organizational conditions. By systematically comparing STEM and non-STEM contexts and incorporating PSs staff alongside academics, the present study addresses these gaps and advances the literature from a descriptive catalogue of concerns towards a structured, multi-level explanation of GenAI adoption in HE. Hypotheses Building on recent scholarship, we contend that GenAI adoption in HE cannot be adequately explained by individual characteristics alone, such as perceived usefulness, ease of use, or technical literacy. Although these factors remain important, accumulating evidence shows that adoption decisions are embedded within wider ethical considerations (Baig & Yadegaridehkordi, 2025), disciplinary traditions, professional norms, and institutions (McGrath et al., 2023; Yan & Qianjun, 2025). These contextual forces shape not only access and capability, but also how staff interpret the risks, responsibilities, and opportunities associated with GenAI. Understanding adoption therefore requires moving beyond a purely cognitive account towards a framework that situates individual perceptions within organizational and disciplinary environments. Disciplinary context is particularly salient in shaping GenAI adoption. Academic fields are structured around distinct epistemic norms, assessment formats, and standards of evidence (Wang et al., 2024). In writing-intensive and interpretive disciplines, concerns about authorship, originality, and academic integrity may be especially pronounced. By contrast, in technically oriented disciplines, issues such as model accuracy, reproducibility, and integration into established workflows may take precedence. Prior research points to cross-disciplinary variation in perceptions of fairness, accountability, and pedagogical alignment (McGrath et al., 2023; Yan & Qianjun, 2025). Although some cross-disciplinary evidence exists for students, including reported differences across STEM and non-STEM cohorts and gendered patterns of attitudes (Arowosegbe et al., 2024; Kim et al., 2025), systematic empirical comparisons among staff remain limited (Zhao et al., 2025). This gap restricts our ability to assess whether disciplinary norms shape barrier perceptions in comparable ways across professional roles within universities. 6 Institutional role constitutes a second structural dimension shaping GenAI adoption. Academic staff are directly responsible for teaching, assessment, and disciplinary knowledge production, whereas PSs and administrative staff operate within governance, compliance, and operational infrastructures. Their exposure to GenAI therefore differs not only in function but also in perceived risk and accountability. PSs staff are more likely to encounter barriers linked to data governance, regulatory compliance, infrastructure, and workflow automation. Academics, by contrast, may foreground concerns about pedagogical integrity, authorship, student capability development, and academic standards. Empirical evidence supports this role-based differentiation. Kutty et al. (2024) demonstrate that institutional position shapes how GenAI is perceived and problematized in HE. Educators tend to frame GenAI primarily in pedagogical and epistemic terms, focusing on assessment integrity, student over reliance, disciplinary standards, and workload implications. Administrators, in contrast, emphasize governance coherence, data privacy, compliance obligations, and institutional risk management. Other studies show that educators, students, and administrators face role-specific barriers, and that professional and administrative staff should be analysed as a distinct group rather than treated as a background category (Hoernig et al., 2024; Hosseini et al., 2023). There is also evidence that non-academic staff are frequently excluded from policy formation processes (Dai et al., 2025). Taken together, this literature suggests that perceived barriers to GenAI adoption are likely to vary systematically between academic and administrative roles, reflecting differences in responsibility, accountability structures, and institutional positioning. The existing literature indicates that institutional position and disciplinary context systematically shape how staff perceive barriers to GenAI adoption. Rather than treating adoption as the outcome of individual attitudes or technical capability alone, recent work suggests that staff interpret risks and opportunities through their location within organizational hierarchies and epistemic communities. These structural positions influence which constraints become salient and which forms of GenAI use appear legitimate, risky, or valuable. To examine this proposition empirically, we formulate the following hypothesis: H1. Perceived barriers to GenAI adoption in HE vary systematically across institutional roles and disciplinary contexts, beyond individual-level attitudes and technical capability. We specify this hypothesis in two dimensions: H1a. Staff working in STEM and non-STEM disciplines will report systematically different GenAI barriers. H1b. Academic staff and PSs staff will exhibit systematically different GenAI barriers. 7 Methodology We examine barriers to GenAI adoption through an in-depth single-institution case study at a Russel Group university. This allows us to analyse variation across disciplinary contexts and institutional roles within a shared governance and policy environment, thereby isolating patterned differences in perceived barriers. We collected data through a cross-sectional survey combining structured quantitative measures with open-ended qualitative responses (DeVellis & Thorpe, 2021; Dillman et al., 2014; Fowler Jr, 2013). Table 1 Questions in the staff survey, where dependent variables have options falling into categories of Ind=Individual, Ins=Institutional, E=Ethical, and C=Cultural. Question Options Note Independent variables In which department or unit do you work at the University? Combo box (dropdown selection plus one line text box) STEM or non- STEM Which of the following best describes your current role? Academic; Professional Services Job Role Dependent variables What are the main barriers to using GenAI in your teaching/work? Lack of institutional guidance or policies (Ins); Concerns about academic integrity and student misuse (E); Limited personal knowledge or training on AI tools (Ind); Lack of time to explore and implement AI solutions (Ind); Ethical concerns about bias, privacy, or surveillance (E); Uncertainty about AIâs effectiveness in improving learning outcomes (Ind); Technical difficulties or lack of institutional support (Ins); Concerns about AI replacing human elements in teaching (E); Resistance from colleagues or institutional culture (C); Lack of funding or access to appropriate AI tools (Ins); Preference for traditional teaching methods (C); Select three most important. Mediator How would you describe your current level of GenAI literacy? Likert scale 0-5, no understanding to expert Literacy What is your overall view of GenAI in education? Likert scale 1-5, very negative to very positive Attitude I am concerned that GenAI will become a threat to my job in the next 5 years. Likert scale 1-5, strongly disagree to strongly agree Job Threat My institution/academic department has provided clear guidance on the acceptable and unacceptable uses of GenAI in teaching/work. Likert scale 1-5, strongly disagree to strongly agree Guidance 8 My institution/academic department has provided sufficient resources to develop staff GenAI literacy. Likert scale 1-5, strongly disagree to strongly agree Support Free-entry texts What are the main barriers to using GenAI in your teaching/work? Free text inputs We developed the survey instrument following the above systematic review of the GenAI literature. The survey was organized around GenAI usage in teaching and professional work, broader perceptions of impact, institutional policy and support, and demographic characteristics. Core barriers in the literature were translated into measurable items capturing institutional role, disciplinary location, GenAI literacy, attitudes towards GenAI, perceived job threat, perceived institutional guidance and support, current and intended GenAI use, and perceived barriers to adoption, where most perceptual variables were measured using five-point Likert scales (DeVellis & Thorpe, 2021) as shown in Table 1. Fig. 1 Conceptual model showing interacting mediators linking independent and dependent variables. To capture barrier structures, respondents selected the three most salient barriers from a predefined list spanning individual, institutional, ethical, and cultural dimensions. This forced-choice design enables us to model barrier profiles as structured outcome categories rather than isolated perceptions. We developed a conceptual model in which independent variables influence a set of mediators, which in turn shape the dependent variables, namely different levels of GenAI adoption barriers, as shown in Fig. 1. Within the mediating layer, we propose that institutional factors shape individual perspectives, including attitudes, AI literacy, and perceived job threat, which in turn influence the level and intensity of perceived barriers. The model is consistent with the previous literature (Gigliotti et al., 2019), and provides a structured framework for examining the mechanisms through which job roles, disciplinary contexts, and broader institutional conditions shape individualsâ perceptions and responses. The model also enables investigation through quantitative approaches such as multinomial logistic regression (MLR) and SEM (Ullman & Bentler, 2012). Independent Variables Dependent Variables Attitude Job Threat Literacy Guidance Support 9 The survey included free-text entries to capture barriers not covered by the predefined items. We analysed these open-ended responses using embedding-based semantic clustering with HDBSCAN (hierarchical density-based spatial clustering of applications with noise) (McInnes et al., 2017). First, we converted the textual responses into high-dimensional vector representations using OpenAI text embedding models. We then applied HDBSCAN to group semantically similar responses. This unsupervised approach allows thematic structures to emerge from the data rather than imposing predefined coding categories. By integrating regression analysis, structural modelling, and semantic clustering within a shared institutional context, we move from identifying patterned differences in barrier profiles to examining the mechanisms that shape them. Results We implemented the survey in Qualtrics using role-based branching to ensure relevance for both academic and PSs staff. Following ethical approval, we administered the survey between mid-June and late July 2025, relying on non-probability recruitment through institutional communications and departmental dissemination. The final dataset comprises 272 valid responses from 72 STEM and 200 non-STEM staff members, of whom 148 were academics and 124 were PSs staff. Barrier distribution by job role and discipline Fig. 2 presents the distribution of barrier categories across disciplinary context and institutional role. The descriptive patterns reveal clear structural clustering. 10 Fig. 2 Barrier percentages differentiating job roles and disciplines shown in a treemap. First, non-STEM teaching staff report ethical barriers more often than other groups. Non-STEM academics account for the largest share of concerns related to academic integrity, over-reliance, and erosion of critical thinking. By contrast, STEM respondents report ethical barriers less frequently. This pattern aligns with disciplinary differences in epistemic norms and assessment formats, where writing-intensive and interpretive disciplines face more immediate challenges to authorship and originality. Second, compared to academic staff, institutional barriers are more prevalent among PSs staff, particularly in non- STEM administrative units. These include concerns related to licensing restrictions, policy ambiguity, governance processes, and infrastructure constraints. This pattern suggests that PSs staff experience GenAI more through organizational systems and compliance structures rather than through classroom pedagogy. Third, individual barriers such as lack of training, limited time, and uncertainty about effectiveness are more evenly distributed across groups but remain discipline sensitive. While present in both STEM and non-STEM contexts, they do not dominate any single structural position. Finally, results show barrier profiles differ markedly by institutional role. Among academic staff, ethical barriers are the most prominent category, especially within non-STEM disciplines, where concerns about academic integrity, student over-reliance, and erosion of critical thinking dominate. In contrast, PSs staff report comparatively fewer ethical concerns and instead display a stronger concentration of individual and institutional barriers, including limitations related to training, time, licensing, infrastructure, and governance. While disciplinary variation persists within both groups, the overall pattern suggests a role-based shift in how GenAI is Ethical Individual Institutional Cultural Academic Professional Services Professional Services Academic Professional Services Academic Academic Professional Services Non - STEM 151 19.7% STEM 40 5.2% Non - STEM 74 9.6% STEM 29 3.8% Non - STEM 94 12.3% STEM 47 6.1% Non - STEM 97 12.6% STEM 33 4.3% Non - STEM 64 8.3% STEM 24 3.1% Non - STEM 38 5.0% STEM 23 3.0% Non - STEM 26 3.4% STEM 6 0.8% Non - STEM 19 2. 5% STEM 2 0. 3% 11 problematized: academics frame GenAI primarily as an epistemic and pedagogical risk, whereas PSs staff experience it more as a capability and organizational implementation challenge. These descriptive differences provide initial support for H1a and H1b: perceived barriers vary systematically across both disciplinary contexts and institutional roles. Multinomial regression of barrier category membership To assess whether the descriptive differences reflect systematic structural effects, we used MLR to predict barrier category membership (Agresti, 2013). We used cultural barriers as the reference category, which allowed comparison with ethical, individual, and institutional barrier categories. The model including STEM status was statistically significant (STEM = 1), as seen in Table 2, indicating that disciplinary affiliation contributes to predicting barrier category membership. However, pseudo-R² values were small (Cox & Snell = .011; Nagelkerke = .012; McFadden = .004), suggesting that discipline explains only a limited proportion of overall variance. Table 2 MLR model fit between STEM/non-STEM and barriers. Model â2 Log Likelihood LR Ď² df p Intercept-only 41.569 Final 33.056 8.512 3 .037 At the category level, compared with cultural barriers, non-STEM respondents had significantly lower odds of being classified into the individual and institutional categories (Table 3). Substantively, this indicates that STEM respondents are more likely than non-STEM respondents to interpret barriers in terms of capability and infrastructure constraints, that is, individual and institutional barriers, rather than as forms of cultural resistance. The contrast for ethical barriers was not statistically significant. Disciplinary context therefore shapes how barriers are interpreted, but its effects are selective rather than uniform across all categories. Table 3 STEM group effects by barrier category (reference outcome = cultural; reference predictor group = STEM) Outcome Category B OR = Exp(B) p Notes Ethical â0.545 0.58 .181 Not significant Individual â0.857 0.424 .035 non-STEM lower odds than STEM Institutional â0.952 0.386 .024 non-STEM lower odds than STEM 12 A second multinomial model examined the effect of job role, with academic staff as the reference group (academic = 1). The model was highly significant (as seen in Table 4), with somewhat larger pseudo-R² values (Cox & Snell = .038; Nagelkerke = .041; McFadden = .016), indicating that institutional role is a stronger structural predictor than discipline, although it still has modest explanatory power. Table 4 MLR model fit between academic/non-academic and barriers. Model â2 Log Likelihood LR Ď² df p Intercept-only 63.516 Final 33.959 29.558 3 <.001 At the category level, PSs staff had significantly higher odds of being classified into the institutional barrier category than into the cultural barriers category (Table 5). Differences between Ethical and Individual barriers were not statistically significant. This result confirms that PSs staff are structurally more likely to frame GenAI constraints in terms of governance, policy clarity, infrastructure, and organizational conditions. Table 5 Job role group effects by barrier category (reference outcome = cultural; reference predictor group = teaching) Outcome category B OR = Exp(B) p Notes Ethical â0.196 0.822 .522 Not significant Individual 0.502 1.653 .101 Not significant Institutional 0.788 2.198 .016 Non-academic higher odds than academic Taken together, the MLR analyses provide qualified support for H1. Both discipline and institutional role significantly predict barrier category membership, confirming that barriers are structured by institutional position. However, the modest explanatory power of the models indicates that structural factors shape barrier framing without fully determining it. The most robust differentiation emerges around institutional barriers, particularly across professional roles, reinforcing the argument that GenAI adoption in HE is embedded in organizational and governance contexts rather than reducible to individual perceptions alone. SEM and hypothesis evaluation SEM was conducted in SPSS Amos 31 using maximum likelihood, with a sample size n = 271 (one incomplete samples was removed). As illustrated in Fig. 1, job role (academic = 1, PSs = 0) and discipline (STEM = 1, non- STEM = 0) were specified as exogenous observed predictors. Four barrier categories were modelled as distinct 13 endogenous variables. Guidance, support, attitude, literacy, and job threat were modelled as intermediate endogenous variables (mediating pathways) linking role/discipline to barrier outcomes. The overall model fit was mixed but acceptable for an exploratory path model, indicated by having Ď² = 31.618 with df = 9 (p < .001), yielding Ď²/df = 3.513. Incremental fit indices were high (CFI = .970, IFI = .972, NFI = .961), and absolute misfit was low (RMR = .034, GFI = .980). However, RMSEA was .096 (90% CI [.061, .134], PCLOSE = .017). Given the low degrees of freedom, the strong CFI/IFI/NFI values indicate that the model captures the covariance structure well. Fig. 3 SEM model structure and path coefficients, where grey indicates non-significant paths. SEM provides pathway-level insight that helps interpret H1 and complements the MLR results. As shown in Fig. 3, discipline (STEM vs non-STEM) significantly predicted attitude (b = .579, SE = .156, CR = 3.722, p < .001), and attitude in turn significantly predicted institutional barriers (b = .189, SE = .034, CR = 5.575, p < .001), indicating an indirect discipline-related pathway to institutional barrier framing. In contrast, role-based differences (academic vs PSs staff) directly predicted institutional barriers (b = â.215, SE = .078, CR = â2.763, p = .006). Taken together, these findings support the claim that barrier framing varies systematically by both disciplinary context and institutional role, beyond individual-level perceptions alone, consistent with H1. At the same time, significant effects were concentrated in selected pathways rather than operating uniformly across the model. SEM also shows that several expected paths were not supported: guidance and support did not significantly predict attitude, literacy, or job threat (all p > .05). However, role significantly predicted literacy (b = .266, SE = .104, CR = 2.549, p = .011), which in turn significantly predicted individual barriers (b = â.260, SE = .055, CR = â4.706, p < .001), indicating an additional indirect role-related pathway. Role also directly predicted ethical barriers (b = .372, SE = .098, CR = 3.786, p < .001), and discipline directly predicted cultural barriers (b = â.121, SE = .057, CR = â2.145, p = .032). Although not all pathways were tested or statistically significant in the MLR analyses, Discipline Literacy Attitude Guidance Cultural Ethical Individual Institutional Job Threat Support Role 14 they provide complementary pathway-level evidence broadly consistent with H1a and H1b, and more importantly, show mediating effects between various factors in GenAI adoption. Free-text cluster themes We first embedded the free-text entries using OpenAIâs text embedding model to represent responses in a high- dimensional semantic vector space. We then applied dimensionality reduction using UMAP (uniform manifold approximation and projection) (Allaoui et al., 2020), followed by HDBSCAN clustering. We tuned the clustering parameters to balance the number of clusters and the distribution of cluster sizes. After forming the clusters, we labelled them using an LLM (gpt-5-mini), and the results are shown in Table 6. From Table 6, it is evident that the free-text clustering reinforces the multi-level structure identified in the quantitative analysis. Three broad thematic domains are observable. Table 6 Top 10 barrier clusters from free-text entries ID Size Description Representative Phrases B1 21 Opposed to GenAI integration 'I will not be integrating GenAI into my teaching', 'I do not intend to integrate GenAI', 'I am strongly against the integration of GenAI', "No, I don't plan to integrate AI into my teaching", 'I have no desire to embed genAI and support a ban' B2 20 Copilot-Only Policy Frustration 'Copilot is the only authorised tool', 'Copilot is inferior to ChatGPT/Claude', 'No approval for competitor products', 'Shifting guidance on data uploads', 'Poor quality and hallucinations' B3 15 Erodes learning and integrity 'Undermines critical thinking and writing', 'Students bypass skills for instant gratification', 'Makes cheating hard to detect; devalues degrees', 'Produces superficially competent but inaccurate work', 'Shortcuts before learning underlying skills' B4 14 Eroding critical thinking 'Students becoming too reliant on GenAI', "GenAI obliterating students' critical-thinking skills", 'Students skip work trusting AI', 'Overreliance on prompting instead of thinking', 'GenAI spreading misinformation as fact' B5 13 Factual Accuracy & Reliability "AI isn't always correct!", 'When AI gets it wrong.', 'factual accuracy in finance', 'still checking the work', 'misinterpreted core message' B6 11 Accuracy and Trust Concerns 'Some staff are mistrustful of it.', 'Not being accurate, so always have to check over what the output is.', 'It is not personal, and content may end up all sounding the same.', "I don't trust it to be either ethical or good enough quality.", 'Whether the information delivered is factual.' B7 11 Institutional access and support barriers 'No access to latest AI tools-subscriptions required', "University won't provide licenses or premium versions", 'Policies make tool approval time- consuming', 'Lack of funding or licensing for AI tools', 'Departmental reluctance and low institutional uptake' B8 9 Skill erosion and assessment validity 'Undermines critical thinking', 'Students over-rely on AI', 'Written essays become invalid', 'Homogenized writing styles', 'AI-generated factual errors' B9 9 Concerns about AI use 'AI encourages shortcut assignments', 'Leads to shallow engagement with tasks', 'Students distrust teachers using AI', 'Teacher dependence may reduce attendance', 'Variability, cost and access issues', 'Students resent AI-driven grading' B10 9 Unclear policy and guidance 'Unclear whether university policy allows it', 'No clear guidance on GDPR and confidentiality', "Don't know who to consult (IDG/legal)", 'Avoid using it due to policy uncertainty', 'Difficulty finding policies on university website' 15 First, a small but distinct group of responses reflects normative resistance (B1), where staff explicitly reject GenAI integration. These statements are not framed in technical or training terms but express principled opposition, aligning with the ethical and cultural barrier categories. Second, several clusters reflect epistemic and pedagogical risk (B3, B4, B8, B9). Respondents emphasize erosion of critical thinking, shortcut learning, declining academic standards, and threats to assessment validity. This language mirrors the ethical barrier category identified in the survey and is consistent with the descriptive finding that academics, particularly in non-STEM disciplines, foreground integrity and skill-development concerns. Third, multiple clusters point to institutional governance and infrastructure constraints (B2, B7, B10), including licensing restrictions, unclear policy guidance, GDPR uncertainty, and limited access to tools. These themes correspond directly to the Institutional barrier category and align with the multinomial finding that PSs staff are more likely to frame barriers organizationally. Finally, clusters focused on accuracy and trust (B5, B6) cut across levels, indicating persistent concerns about reliability and output quality. Taken together, the qualitative data substantiates the structural differentiation identified quantitatively: while GenAI use is widespread, the perceived barriers are shaped by disciplinary norms, professional role, and governance context rather than by individual reluctance alone. Discussions Adoption beyond the individual level The findings support growing arguments that technology adoption in AI-enabled environments cannot be understood through individual-level acceptance models alone (FakhrHosseini et al., 2024). While constructs such as perceived usefulness and ease of use remain relevant, they do not fully capture how adoption unfolds in HE, where technologies are embedded within assessment regimes, governance frameworks, professional norms, and regulatory constraints. Across analytical stages, a consistent pattern emerges. The descriptive results show that barriers cluster in structured ways rather than appearing as evenly distributed individual concerns. The MLR results confirm that barrier categories are systematically associated with institutional position, even if the explanatory power of single structural variables remains modest. The SEM extends this insight by demonstrating differentiated pathways: 16 literacy strongly predicts capability-based barriers, ethical concerns are linked to perceptions of professional vulnerability, and institutional barriers are associated with governance and organizational conditions. Finally, the text clustering analysis illustrates how these dynamics are expressed in practice, with respondents framing barriers in terms of assessment integrity, policy ambiguity, infrastructure constraints, reliability concerns, and normative resistance. Taken together, these findings indicate that GenAI adoption in HE emerges from the interaction of user competence, task demands, professional identity, and institutional governance. Individual skills and attitudes matter, but they operate within broader structural conditions that shape how usefulness, risk, and responsibility are interpreted. Addressing adoption challenges therefore requires attention not only to training and confidence- building, but also to the institutional and normative environments in which GenAI is deployed. PSs as a distinct site of GenAI adoption Another key contribution of the current study is the inclusion of PSs staff alongside academics. While recent studies increasingly recognize that GenAI adoption in HE extends beyond teaching and learning (An et al., 2025; Jin et al., 2025), empirical research still tends to focus on the classroom contexts and aggregate PSs roles into âadministratorâ categories (Kutty et al., 2024). This underplays the important functions PSs staff perform in enabling institutions to operate, such as governance, compliance, and student services etc. By systematically comparing academic and PSs staff within the same institutional context, we address a significant gap in the literature and provide a structured analysis of both the similarities and differences between the two groups. Our results suggest that PSs staff are more likely to report institutional barriers, and their institutional role has a direct association with this barrier category. We also found that, compared with academics, PSs staff also frame the problem differently, emphasizing licensing restrictions, policy ambiguity, and governance processes. This pattern is consistent with emerging evidence from the literature (ChristâBrendemĂźhl, 2025; Dai et al., 2025), but our findings extend this work by showing that HE institutions face multiple adoption problems simultaneously: a pedagogical adoption problem in academic domains such as academic integrity and student learning experience and outcomes, and an organizational adoption problem in PSs domains where challenges are more strongly linked to governance and implementation challenge embedded in institutional systems and accountability structures. The findings have practical significance for institutional strategy. First, role-specific implementation guidance is required, particularly around tool approval, procurement processes, and operational use cases. This is important not only for addressing current barriers such as the âCopilot-only policyâ (e.g. B2 in Table 6), but also for helping institutions remain adaptive in a rapidly evolving GenAI ecosystem. Second, our results support recent calls to 17 widen GenAI adoption research beyond pedagogy and to examine the full institutional ecology (FakhrHosseini et al., 2024; Shrivastava, 2025). In this respect, PSs staff should not be treated as a background category, but as a distinct analytical group whose experiences are essential for understanding GenAI adoption in HE. Gains and trade-offs of the multi-method design The current study employs a multi-method analytical design based on the data collected from a staff survey. This allows us to move beyond simple descriptive accounts and toward a more structured explanation and comparison (Creswell & Clark, 2007). Recent progress in the literature highlights existing work in GenAI adoption (Ayanwale, 2024; Nikolic et al., 2024; Shrivastava, 2025), however, descriptive statistics or single-model acceptance frameworks often fail to address relational dynamics through which adoption unfolds in complex institutional settings. Our analytical pipeline takes advantage of and combines several approaches: descriptive analyses establish empirical patterns; MLR and SEM test statistical significance and estimate pathway-level relationships among roles, disciplines, and mediating variables; and clustering of free-text responses captures themes and provides context that is difficult to fully represent in predefined survey items. Taken together, this multi-method design provides a richer account of GenAI adoption than would be possible through any single approach. However, there are certain caveats. First, MLR showed modest pseudo-R² values, indicating that role and discipline are meaningful but incomplete predictors of barrier category membership. Second, SEM improves purely descriptive analyses, however, causal claims such as from roles to different barriers should be made cautiously. At present, the evidence supports association rather than causal effect. Third, free-text clustering and LLM-assisted labelling provide scalable thematic analysis, but they involve parameter fine-tuning that can substantially influence cluster granularity and interpretation (Allaoui et al., 2020; McInnes et al., 2017). Conclusion This study investigated the barriers to GenAI adoption at different levels through a multi-method survey of 272 academic and PSs staff at a Russell Group university. By integrating MLR, SEM, and semantic clustering, we demonstrated that adoption barriers are not merely individual but are systematically shaped by institutional roles and disciplinary contexts. Our findings confirm that non-STEM academics primarily face ethical and pedagogical concerns, whereas STEM and PSs staff disproportionately encounter institutional and governance constraints. This research makes three contributions with significant implications for university policy and technology adoption frameworks. First, by providing a systematic cross-disciplinary comparison within a single institutional 18 environment, we highlight how distinct epistemic norms dictate the risks of GenAI, suggesting that generic, university-wide adoption policies are likely to be ineffective. Second, extending the analytical focus beyond teaching academics to include PSs staff uncovers a critical yet frequently overlooked dimension of organizational adoption centred on compliance, licensing, and student services. Third, our methodological integration of quantitative modelling with LLM-assisted qualitative clustering advances the field past descriptive analyses toward multi-level, structural explanations. Consequently, HE institutions can leverage these insights to design targeted, role-specific governance processes, frameworks, and strategies that directly address the localized adoption barriers for their diverse workforce. In addition, this study presents opportunities for future research. The data relies on a cross-sectional, non- probability sample from a single institution. While SEM highlighted key associations, it cannot definitively establish causal relationships, for example, between institutional positions and specific adoption barriers. Additionally, the modest explanatory power of our regression models suggests that other unmeasured variables may also play pivotal roles. Future longitudinal studies across diverse institutional types, coupled with in-depth qualitative interviews, will be necessary to validate these structural pathways and track how GenAI barrier perceptions evolve as institutional policies mature. Data availability statement The data that support the findings of this study are available from the corresponding author upon reasonable request, subject to non-disclosure requirements. Funding statement The authors wish to thank Warwick Education Fund (project âIdentifying Barriers and Use Cases for Generative AI in Education Using Retrieval-Augmented Generation and Staff Surveysâ) for financial support. Conflict of interest disclosure The authors disclose that they have no actual or perceived conflicts of interest. 19 Ethics approval statement Ethical approval for this study was granted by Warwickâs Biomedical and Scientific Research Ethics Committee (BSREC) on 25 February 2025, under reference number BSREC 85/24-25. References Abbas, M., Jam, F. A., & Khan, T. I. (2024). Is it harmful or helpful? Examining the causes and consequences of generative AI usage among university students. International Journal of Educational Technology in Higher Education, 21(1), 10. Agresti, A. (2013). Categorical data analysis. John Wiley & Sons. Al-Abdullatif, A. M. (2024). Modeling teachersâ acceptance of generative artificial intelligence use in higher education: The role of AI literacy, intelligent TPACK, and perceived trust. Education Sciences, 14(11), 1209. Al-Samarraie, H., Sarsam, S. M., Ibrahim Alzahrani, A., Chatterjee, A., & Swinnerton, B. J. (2025). Gender perceptions of generative AI in higher education. Journal of Applied Research in Higher Education, 17(5), 1944-1958. Al-Shabandar, R., Jaddoa, A., Elwi, T. A., Mohammed, A., & Hussain, A. J. (2024). A systematic review for the implication of generative AI in higher education. Infocommunications Journal, 16(3), 31-42. Allaoui, M., Kherfi, M. L., & Cheriet, A. (2020). Considerably improving clustering algorithms using UMAP dimensionality reduction technique: a comparative study. International conference on image and signal processing, An, Y., Yu, J. H., & James, S. (2025). Investigating the higher education institutionsâ guidelines and policies regarding the use of generative AI in teaching, learning, research, and administration. International Journal of Educational Technology in Higher Education, 22(1), 10. Ardito, C. G. (2025). Generative AI detection in higher education assessments. New Directions for Teaching and Learning, 2025(182), 11-28. Arowosegbe, A., Alqahtani, J. S., & Oyelade, T. (2024). Perception of generative AI use in UK higher education. Frontiers in Education, Asad, M. M., & Ajaz, A. (2024). Impact of ChatGPT and generative AI on lifelong learning and upskilling learners in higher education: unveiling the challenges and opportunities globally. The International Journal of Information and Learning Technology, 41(5), 507-523. Ayanwale, M. A. (2024). Using diffusion theory of innovation to investigate perceptions of STEM and non-STEM students' adoption of chatbot systems in higher education: A multiple group analysis. 2024 IEEE Global Engineering Education Conference (EDUCON), Baig, M. I., & Yadegaridehkordi, E. (2025). Factors influencing academic staff satisfaction and continuous usage of generative artificial intelligence (GenAI) in higher education. International Journal of Educational Technology in Higher Education, 22(1), 5. Batista, J., Mesquita, A., & Carnaz, G. (2024). Generative AI and higher education: Trends, challenges, and future directions from a systematic literature review. Information, 15(11), 676. Bearman, M., Tai, J., Dawson, P., Boud, D., & Ajjawi, R. (2024). Developing evaluative judgement for a time of generative artificial intelligence. Assessment & Evaluation in Higher Education, 49(6), 893-905. Cabero-Almenara, J., Palacios-RodrĂguez, A., Loaiza-Aguirre, M. I., & Andrade-Abarca, P. S. (2024). The impact of pedagogical beliefs on the adoption of generative AI in higher education: predictive model from UTAUT2. Frontiers in Artificial Intelligence, 7, 1497705. Chan, C. K. Y., & Hu, W. (2023). Studentsâ voices on generative AI: Perceptions, benefits, and challenges in higher education. International Journal of Educational Technology in Higher Education, 20(1), 43. Chiu, T. K. (2024). The impact of Generative AI (GenAI) on practices, policies and research direction in education: A case of ChatGPT and Midjourney. Interactive Learning Environments, 32(10), 6187-6203. ChristâBrendemĂźhl, S. (2025). Leveraging generative AI in higher education: An analysis of opportunities and challenges addressed in university guidelines. European Journal of Education, 60(1), e12891. CoffĂŠ, H., Crawley, S., & Givens, J. (2026). Growing polarisation: ideology and attitudes towards climate change. West European Politics, 49(01), 1-29. Creswell, J. W., & Clark, V. P. (2007). Mixed methods research. Thousand Oaks, CA. 20 Dabis, A., & CsĂĄki, C. (2024). AI and ethics: Investigating the first policy responses of higher education institutions to the challenge of generative AI. Humanities and Social Sciences Communications, 11(1), 1-13. Dai, Y., Lai, S., Lim, C. P., & Liu, A. (2025). University policies on generative AI in Asia: Promising practices, gaps, and future directions. Journal of Asian Public Policy, 18(2), 260-281. Daniel, K., Msambwa, M. M., & Wen, Z. (2025). Can generative AI revolutionise academic skills development in higher education? A systematic literature review. European Journal of Education, 60(1), e70036. Davis, F. D. (1989). Perceived Usefulness, Perceived Ease of Use and User Acceptance of Information Technology. MIS quarterly. DeVellis, R. F., & Thorpe, C. T. (2021). Scale development: Theory and applications. Sage publications. Dillman, D. A., Smyth, J. D., & Christian, L. M. (2014). Internet, phone, mail, and mixed-mode surveys: The tailored design method. Indianapolis, Indiana, 17. Essien, A., Salami, A., Ajala, O., Adebisi, B., Shodiya, A., & Essien, G. (2024). Exploring socio-cultural influences on generative AI engagement in Nigerian higher education: an activity theory analysis. Smart learning environments, 11(1), 63. Evangelista, E. D. L. (2025). Ensuring academic integrity in the age of ChatGPT: Rethinking exam design, assessment strategies, and ethical AI policies in higher education. Contemporary Educational Technology, 17(1), ep559. FakhrHosseini, S., Chan, K., Lee, C., Jeon, M., Son, H., Rudnik, J., & Coughlin, J. (2024). User adoption of intelligent environments: A review of technology adoption models, challenges, and prospects. International Journal of HumanâComputer Interaction, 40(4), 986-998. Fowler Jr, F. J. (2013). Survey research methods. Sage publications. Gasaymeh, A.-M. M., Beirat, M. A., & Abu Qbeita, A. a. A. (2024). University studentsâ insights of generative artificial intelligence (AI) writing tools. Education Sciences, 14(10), 1062. Ghimire, A., & Edwards, J. (2024). Generative AI adoption in classroom in context of technology acceptance model (TAM) and the innovation diffusion theory (IDT). arXiv preprint arXiv:2406.15360. Gigliotti, R., Vardaman, J., Marshall, D. R., & Gonzalez, K. (2019). The role of perceived organizational support in individual change readiness. Journal of Change Management, 19(2), 86-100. Hoernig, S., Ilharco, A., Pereira, P. T., & Pereira, R. (2024). Generative AI and higher education: Challenges and opportunities. Institute of Public Policy, 12(2), 45-60. Hosseini, M., Gao, C. A., Liebovitz, D. M., Carvalho, A. M., Ahmad, F. S., Luo, Y., MacDonald, N., Holmes, K. L., & Kho, A. (2023). An exploratory survey about using ChatGPT in education, healthcare, and research. Plos one, 18(10), e0292216. Jayaram, Y., Sundar, D., & Bhat, J. (2024). Generative AI Governance & Secure Content Automation in Higher Education. International Journal of Artificial Intelligence, Data Science, and Machine Learning, 5(4), 163-174. Jin, Y., Yan, L., Echeverria, V., GaĹĄeviÄ, D., & Martinez-Maldonado, R. (2025). Generative AI in higher education: A global perspective of institutional adoption policies and guidelines. Computers and Education: Artificial Intelligence, 8, 100348. Khan, S. (2024). Brave new words: How AI will revolutionize education (and why that's a good thing). Penguin. Khlaif, Z. N., Ayyoub, A., Hamamra, B., Bensalem, E., Mitwally, M. A. A., Ayyoub, A., Hattab, M. K., & Shadid, F. (2024). University Teachersâ Views on the Adoption and Integration of Generative AI Tools for Student Assessment in Higher Education. Education Sciences, 14(10), 1090. https://w.mdpi.com/2227-7102/14/10/1090 Kim, J., Klopfer, M., Grohs, J. R., Eldardiry, H., Weichert, J., Cox, L. A., & Pike, D. (2025). Examining faculty and student perceptions of generative AI in university courses. Innovative Higher Education, 50(4), 1281-1313. Kutty, S., Chugh, R., Perera, P., Neupane, A., Jha, M., Li, L., Gunathilake, W., & Perera, N. C. (2024). Generative AI in higher education: Perspectives of students, educators and administrators. Journal of Applied Learning & Teaching, 7(2), 47-60. Lan, G., Feng, X., Du, S., Song, F., & Xiao, Q. (2025). Integrating ethical knowledge in generative AI education: Constructing the GenAI-TPACK framework for university teachersâ professional development. Education and Information Technologies, 30(11), 15621-15644. Lee, V. R., Parli, V., Hau, I., Hynes, P., & Zhang, D. (2025). Response to the Department of Educationâs Request for Information on AI in Education. Stanford University Human-Centered Artificial Intelligence. Retrieved 26 December 2025 from https://hai.stanford.edu/assets/files/hai-stanford-accelerator-for- learning-rfi-response-advancing-ai-in-education.pdf Lindebaum, D., & Fleming, P. (2024). ChatGPT undermines human reflexivity, scientific responsibility and responsible management research. British Journal of Management, 35(2), 566-575. 21 Mah, D.-K., & GroĂ, N. (2024). Artificial intelligence in higher education: exploring faculty use, self-efficacy, distinct profiles, and professional development needs. International Journal of Educational Technology in Higher Education, 21(1), 58. McGrath, C., Pargman, T. C., Juth, N., & Palmgren, P. J. (2023). University teachers' perceptions of responsibility and artificial intelligence in higher education-An experimental philosophical study. Computers and Education: Artificial Intelligence, 4, 100139. McInnes, L., Healy, J., & Astels, S. (2017). hdbscan: Hierarchical density based clustering. J. Open Source Softw., 2(11), 205. Michel-Villarreal, R., Vilalta-Perdomo, E., Salinas-Navarro, D. E., Thierry-Aguilera, R., & Gerardou, F. S. (2023). Challenges and opportunities of generative AI for higher education as explained by ChatGPT. Education Sciences, 13(9), 856. Nguyen, K. V. (2025). The use of generative AI tools in higher education: Ethical and pedagogical principles. Journal of Academic Ethics, 23(3), 1435-1455. Nikolic, S., Wentworth, I., Sheridan, L., Moss, S., Duursma, E., Jones, R. A., Ros, M., & Middleton, R. (2024). A systematic literature review of attitudes, intentions and behaviours of teaching academics pertaining to AI and generative AI (GenAI) in higher education: An analysis of GenAI adoption using the UTAUT framework. Australasian Journal of Educational Technology, 40(6), 56-75. OâDea, X. (2024). Generative AI: is it a paradigm shift for higher education? Studies in higher education, 49(5), 811-816. Shata, A. (2025). âOpting Out of AIâ: exploring perceptions, reasons, and concerns behind faculty resistance to generative AI. Frontiers in Communication, 10, 1614804. Shrivastava, P. (2025). Understanding acceptance and resistance toward generative AI technologies: a multi- theoretical framework integrating functional, risk, and sociolegal factors. Frontiers in Artificial Intelligence, 8, 1565927. Strzelecki, A., & ElArabawy, S. (2024). Investigation of the moderation effect of gender and study level on the acceptance and use of generative AI by higher education students: Comparative evidence from Poland and Egypt. British Journal of Educational Technology, 55(3), 1209-1230. Sullivan, M., Kelly, A., & McLaughlan, P. (2023). ChatGPT in higher education: Considerations for academic integrity and student learning. Journal of Applied Learning & Teaching, 6(1), 31-40. ThĂźs, D., Malone, S., & BrĂźnken, R. (2024). Exploring generative AI in higher education: a RAG system to enhance student engagement with scientific literature. Frontiers in Psychology, 15, 1474892. Ullman, J. B., & Bentler, P. M. (2012). Structural equation modeling. Handbook of psychology, second edition, 2. Valdivieso, T., & GonzĂĄlez, O. (2025). Generative AI tools in Salvadoran higher education: Balancing equity, ethics, and knowledge management in the global south. Education Sciences, 15(2), 214. Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view1. MIS quarterly, 27(3), 425-478. Wang, H., Dang, A., Wu, Z., & Mac, S. (2024). Generative AI in higher education: Seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence, 7, 100326. Wang, T., Lund, B. D., Marengo, A., Pagano, A., Mannuru, N. R., Teel, Z. A., & Pange, J. (2023). Exploring the potential impact of artificial intelligence (AI) on international students in higher education: Generative AI, chatbots, analytics, and international student success. Applied sciences, 13(11), 6716. Watermeyer, R., Lanclos, D., Phipps, L., Shapiro, H., Guizzo, D., & Knight, C. (2025). Academicsâ Weak (ening) resistance to generative AI: The cause and cost of prestige? Postdigital Science and Education, 7(4), 1171-1191. Williams, M. D., Rana, N. P., & Dwivedi, Y. K. (2015). The unified theory of acceptance and use of technology (UTAUT): a literature review. Journal of enterprise information management, 28(3), 443-488. Yan, Z., & Qianjun, T. (2025). Integrating AI-generated content tools in higher education: a comparative analysis of interdisciplinary learning outcomes. Scientific Reports, 15(1), 25802. Yusuf, A., Pervin, N., & RomĂĄn-GonzĂĄlez, M. (2024). Generative AI and the future of higher education: a threat to academic integrity or reformation? Evidence from multicultural perspectives. International Journal of Educational Technology in Higher Education, 21(1), 21. Zhao, X., Liu, C., Philippakos, Z., Zahra, F., & Aydeniz, M. (2025). Reflections on the merit and perils of AI in higher education: Five early adopterâs perspectives. International Journal of Technology in Education and Science, 9(4), 522-544.