The Unified Theory of Acceptance and Use of Technology (UTAUT), developed by Venkatesh et al. (2003), integrates multiple technology acceptance models into four core predictors of technology adoption: performance expectancy, effort expectancy, social influence, and facilitating conditions. In higher education, these constructs explain how teachers evaluate the functional usefulness, ease of use, social legitimacy, and institutional support of emerging technologies such as generative artificial intelligence (GAI).

Empirical studies consistently demonstrate UTAUT’s strength across educational contexts. Manswar et al. (2024) synthesized two decades of UTAUT-based research in education and confirmed its sustained explanatory power amid the expansion of digital and AI-assisted learning. Feng and Xiaoyu (2024) reported that performance expectancy, effort expectancy, and social influence significantly predicted AI-related acceptance in Malaysian humanities education, while facilitating conditions exerted weaker but still relevant effects. Similarly, Julio (2024) found that all four UTAUT constructs predicted AI adoption among university teachers, with pedagogical beliefs further strengthening these relationships. Xinhua and Thitinant (2024) observed that teaching experience and IT proficiency moderated UTAUT effects among preschool teachers, underscoring the framework’s sensitivity to contextual and demographic factors. In English-language teaching contexts, Anabel et al. (2024) noted that social norms and personal attitudes outweighed infrastructure constraints, while Lanying and Chunming (2023) demonstrated that UTAUT predicts not only behavioral intention but also affective outcomes such as learner satisfaction.

Although UTAUT has been extensively validated, most prior studies treat its constructs as direct predictors of behavioral intention, with limited attention to the motivational mechanisms through which these enablers shape teachers’ evaluative judgments. Moreover, UTAUT research on AI adoption has largely focused on technologically advanced or monolingual contexts, leaving regional, multilingual English-language teaching environments underexplored. In Philippine higher education, particularly in Cebu, where institutional resources and professional norms vary across urban and rural campuses, there is a need to examine how UTAUT-based enablers indirectly influence adoption through teachers’ value-based appraisals rather than solely through direct effects.

Expectancy–Value Theory (EVT) complements UTAUT by emphasizing internal motivational processes, proposing that individuals’ behavior is shaped by their expectations for success and the subjective value they assign to tasks (Goegan et al., 2021). In instructional contexts, this implies that teachers adopt generative AI not solely because it is usable or supported, but because it aligns with their professional goals, interests, and perceived costs.

Empirical evidence supports the relevance of EVT in educational technology adoption. Durk et al. (2024) demonstrated that female physics lecturers sustained technology use when they perceived strong educational value, even in the presence of challenges. Wong and Chapman (2024) further strengthened EVT’s applicability by developing the Expectancy–Value Higher Education Instrument (EVHEI), which reliably measures expectancy and task value constructs in academic settings. In AI-specific contexts, Prema et al. (2023) found that perceived task value strongly predicted students’ intention to use ChatGPT, although existing studies have focused predominantly on learners rather than educators.

Despite EVT’s explanatory potential, few studies have systematically applied its four core value components, attainment value, intrinsic value, utility value, and perceived cost, to teachers’ adoption of generative AI. More importantly, EVT is rarely integrated with UTAUT in a way that theoretically distinguishes external enablers from internal motivational appraisals. This study addresses these gaps by positioning EVT constructs as mediating mechanisms through which UTAUT-based enablers influence behavioral intention, thereby extending EVT to professional ELT contexts and strengthening theoretical integration.

To capture the confidence-based and ethical dimensions of AI adoption, this study extends UTAUT and EVT by incorporating perceived knowledge of AI (PKA) and perceived privacy concerns (PPC) as distinct explanatory constructs. Perceived knowledge of AI (PKA) is operationally defined as teachers’ subjective evaluation of their own understanding, familiarity, and perceived capability to use generative AI tools for instructional purposes. This construct reflects self-assessed confidence in navigating AI applications, interpreting outputs, and applying them in teaching contexts, rather than objectively measured technical proficiency or formally assessed competence. Prior research indicates that higher levels of perceived knowledge increase educators’ likelihood of experimenting with and adopting AI technologies (Yue et al., 2024; Oran, 2023).

Perceived privacy concerns (PPC) are operationalized as teachers’ anticipated perceptions of ethical, legal, and data-related risks associated with the use of generative AI in educational settings (Alashoor et al., 2023). These concerns include apprehensions regarding student data protection, confidentiality, informed consent, and compliance with institutional and national data privacy regulations. In contexts with developing or unevenly enforced data governance frameworks, such as the Philippines, heightened privacy concerns may constrain teachers’ willingness to engage with AI technologies (Gella, 2024). Importantly, PPC represents risk perception rather than actual policy knowledge, capturing teachers’ subjective sensitivity to potential ethical implications of AI use.

Motivational factors interact with these constructs. Teachers who perceive high utility value in AI tools are more inclined to develop AI-related competence, thereby strengthening perceived knowledge (Ibrahim and Shiring, 2022). Conversely, high perceived cost, conceptualized as a negative task valuation encompassing time demands, effort, and perceived professional or ethical risk, may heighten perceived privacy concerns, particularly in settings where institutional safeguards and clear guidelines are limited (Mahama et al., 2024; Wagman et al., 2023).

Performance expectancy (PE) is a key construct in the Unified Theory of Acceptance and Use of Technology (UTAUT), referring to the extent to which teachers believe generative AI will enhance their teaching effectiveness (Venkatesh et al., 2003). In higher education, this relates directly to the perceived attainment value, which is the belief that AI use leads to meaningful professional outcomes (Chiu and Wang, 2008). Teachers who anticipate benefits such as automated assessment, personalized feedback, and efficient content delivery are more likely to view AI as valuable in achieving instructional goals (Chan and Tsi, 2024; Rajak et al., 2024). However, this relationship is influenced by factors such as institutional support, training, and alignment with pedagogical goals (Wang et al., 2021). Therefore, performance expectancy can only translate into perceived attainment value when barriers like limited AI literacy and ethical concerns are addressed (Pedro et al., 2019). Thus, this study has a hypothesis that:

Effort expectancy (EE) influences perceived utility value (UV) in instructional practice (Venkatesh et al., 2003). Teachers are more likely to perceive generative AI as useful when it is accessible, intuitive, and easy to integrate into lesson planning and assessment tasks (Cabero-Almenara et al., 2024; Wu et al., 2022). Simplified interfaces and responsive institutional support further enhance willingness to engage with AI tools, particularly in resource-limited settings (Manswar et al., 2024). Conversely, perceptions of technological complexity have been associated with less favorable evaluations of instructional technologies, which may attenuate perceived utility (Anabel et al., 2024). Thus, user-friendly design and targeted training are expected to enhance perceived utility. This paper hypothesizes that:

Within the Unified Theory of Acceptance and Use of Technology, social influence (SI) reflects the extent to which significant others, such as colleagues, institutional leaders, and professional communities, shape individuals’ evaluations of a technology’s legitimacy and relevance. However, rather than exerting a direct effect on behavioral intention, social influence is theorized to operate indirectly by shaping teachers’ value-based appraisals, as conceptualized in Expectancy–Value Theory. In collectivist-oriented professional environments, normative endorsement and peer modeling contribute to the internalization of shared goals and standards (Hennessy et al., 2022), thereby increasing the perceived importance (attainment value) and enjoyment (intrinsic value) associated with adopting generative AI (Lan, 2024). When respected colleagues or administrators frame AI use as aligned with professional expectations or pedagogical excellence, teachers are more likely to see AI as meaningful to their roles and identities as educators.

Furthermore, institutional advocacy and leadership support signal the instrumental relevance of generative AI for accomplishing instructional tasks, thereby strengthening perceived utility value (UV) (Chounta et al., 2022). Although Chounta et al. (2022) examined AI perceptions in K–12 education, the study underscores how institutional norms and leadership endorsement influence educators’ evaluations of emerging technologies, a mechanism also relevant to higher education ELT. Through these mechanisms, social influence does not merely encourage compliance but reshapes how teachers evaluate the professional worth, practical usefulness, and experiential appeal of AI technologies. Consequently, social influence is expected to affect behavioral intention indirectly through its impact on attainment value, intrinsic value, and utility value, rather than through a direct pathway. Accordingly, this study hypothesizes that:

Facilitating conditions (FC), including infrastructure, technical support, and ongoing training, enable teachers to use generative AI effectively (Ibrahim and Shiring, 2022). Although Ibrahim and Shiring (2022) adopt a TAM-based framework, their findings on perceived usefulness and support structures conceptually align with facilitating conditions and utility value as modeled in UTAUT and EVT. These factors significantly affect perceived utility value (UV) and perceived cost (PC). Teachers are more likely to find AI beneficial when institutions provide structured support for implementation (Hojeij et al., 2024). Findings from emergency remote learning contexts (Wagman et al., 2023) suggest that limited support can amplify perceptions of technical burden, a concern that may also arise in AI adoption within higher education. Well-supported environments reduce perceived barriers and increase the likelihood of integration. This hypothesizes that:

Within Expectancy–Value Theory, perceived attainment value (AV) reflects the degree to which engaging in a task is considered important for one’s professional identity, goals, and sense of accomplishment. When teachers perceive the use of generative AI as instrumental to their effectiveness, professional growth, or instructional relevance, they are more likely to invest cognitive and motivational resources in learning how the technology works. This value-driven investment encourages deliberate effort, sustained engagement, and self-directed exploration, which in turn enhances teachers’ perceived knowledge of AI (PKA), defined as their self-assessed confidence and understanding of AI tools. Prior research supports this mechanism, showing that educators who assign high professional importance to instructional technologies are more inclined to pursue skill development and instructional innovation (Alegre, 2023; Assassi, 2025). Accordingly, this study hypothesizes that:

Perceived intrinsic value (IV), or enjoyment from using AI, encourages sustained engagement and skill development. Teachers who find AI intellectually stimulating are more willing to explore it further and apply it creatively in teaching (Sharma and Srivastava, 2020; Zimmerman, 2018). This curiosity leads to a deeper understanding and broader application in their perceived knowledge of AI (PKA). Hence, this paper hypothesizes that:

Perceived utility value (UV) is defined as teachers’ evaluation of generative AI as instrumentally useful and efficient for accomplishing instructional tasks, such as providing feedback, generating learning materials, or supporting assessment. Teachers who perceive high utility value are more motivated to invest effort in learning and experimenting with AI tools, which strengthens their perceived knowledge of AI (PKA) and self-assessed competence (Mah and Groß, 2024). At the same time, increased engagement with AI systems may heighten teachers’ awareness of ethical and data-related risks, thereby intensifying their perceived privacy concerns (PPC) (Popenici and Kerr, 2017). This dual influence positions utility value as a mechanism that simultaneously promotes technical confidence and ethical sensitivity. Accordingly, this study proposes the following hypotheses:

Perceived cost (PC) is operationalized as a negative task valuation reflecting teachers’ perceptions of the time investment, effort, cognitive load, and potential ethical or professional risks associated with the use of generative AI in instructional practice. When perceived cost is high, educators may become more attentive to possible data protection and ethical compliance issues, thereby intensifying their perceived privacy concerns (PPC) (Mahama et al., 2024; Wagman et al., 2023). This relationship is particularly salient in under-resourced educational settings, where limited institutional safeguards and unclear implementation guidelines may amplify risk sensitivity. Accordingly, this study theorizes that:

Perceived knowledge of AI (PKA) is operationalized as teachers’ subjective self-assessment of their understanding, familiarity, and perceived capability (Gado et al., 2021), to use generative AI tools for instructional purposes, rather than as objectively measured technical competence. This construct reflects educators’ confidence in navigating AI applications, interpreting AI-generated outputs, and applying them appropriately in teaching and assessment contexts. Teachers with higher perceived knowledge are more likely to integrate AI into their instructional practices and express stronger behavioral intention (BI) to use AI tools (Oran, 2023). Professional training, hands-on exposure, and guided practice play a critical role in strengthening perceived knowledge, thereby facilitating adoption (Shao et al., 2024). Accordingly, this study assumes that:

Perceived privacy concerns (PPC) are operationalized as teachers’ subjective perceptions of ethical, legal, and data-related risks associated with the use of AI technologies in instructional contexts, including concerns about student data protection, confidentiality, consent, and regulatory compliance (Alashoor et al., 2023). Even when generative AI is perceived as pedagogically beneficial, heightened privacy concerns may reduce teachers’ willingness to use such technologies if adequate safeguards and clear guidelines are perceived to be lacking (Aldboush and Ferdous, 2023). In the Philippine higher education context, where awareness of data privacy regulations varies across institutions, alignment with national privacy laws and institutional policies is critical for fostering trust and supporting responsible AI adoption (Gella, 2024). Accordingly, this study assumes that:

Age is examined as a moderating variable that may influence the strength of relationships among the key constructs in the proposed model. Prior research suggests that age shapes how educators evaluate technology usability, respond to social support, and develop confidence in using digital tools (Morris and Venkatesh, 2000; Tewathia et al., 2020). Differences in professional experience, prior exposure to educational technologies, and learning preferences may lead teachers of different age groups to interpret the benefits, costs, and risks of generative AI differently. Accordingly, age is modeled as a moderator across the structural relationships specified in this study. Thus, this paper hypothesizes that age moderates the relationships among the constructs in the proposed model.

Sex is also examined as a moderating variable in the proposed model. Prior research indicates that male and female educators may differ in how they evaluate technology-related factors such as performance expectancy, effort expectancy, and social influence, reflecting variations in socialization, professional experiences, and institutional contexts (Ong and Lai, 2006; Arthur et al., 2024). Rather than assuming inherent differences in adoption behavior, this study treats sex as a contextual moderator that may influence the strength of relationships among the model constructs. Accordingly, sex is modeled as a moderator across all structural relationships specified in this study. Thus, this paper further proposes that sex moderates the relationships among the constructs in the proposed model.

By integrating UTAUT, EVT, perceived knowledge of AI, perceived privacy concerns, and the moderating factors of age and sex, this study proposes a comprehensive and a novel model to understand behavioral intention to use generative AI among higher education teachers in the Philippines (see Figure 1).

Figure 1 illustrates the proposed structural model integrating the Unified Theory of Acceptance and Use of Technology (UTAUT), Expectancy–Value Theory (EVT), perceived knowledge of AI, perceived privacy concerns, and demographic moderators to explain higher education teachers’ behavioral intention to use generative AI. The model is organized into three conceptual layers. The first layer consists of UTAUT-based external enablers, including performance expectancy (PE), effort expectancy (EE), social influence (SI), and facilitating conditions (FC). These constructs do not directly predict behavioral intention but are theorized to shape teachers’ value-based motivational appraisals. The second layer comprises EVT constructs, namely perceived attainment value (AV), perceived intrinsic value (IV), perceived utility value (UV), and perceived cost (PC), which capture teachers’ evaluative judgments regarding the importance, enjoyment, usefulness, and negative trade-offs associated with generative AI use. In the third layer, perceived knowledge of AI (PKA) and perceived privacy concerns (PPC) represent downstream cognitive and ethical mechanisms linking value appraisals to behavioral intention to use AI (BI). Solid arrows indicate hypothesized direct effects, while dashed arrows represent moderation effects. Age and sex are modeled as moderators influencing all structural relationships, reflecting potential demographic differences in how teachers evaluate and respond to behavioral intention toward generative AI adoption. Overall, the model specifies an indirect, theory-driven pathway in which external enablers influence behavioral intention primarily through motivational, cognitive, and ethical processes rather than through direct effects alone.

This study employed a quantitative, non-experimental research design, specifically a predictive correlational design, to examine the structural relationships among latent variables derived from the Unified Theory of Acceptance and Use of Technology (UTAUT), Expectancy–Value Theory (EVT), and additional constructs, such as perceived knowledge of AI (PKA) and perceived privacy concerns (PPC).

The study is explicitly predictive in orientation with theory-guided testing, aiming to explain and estimate the strength of relationships among constructs rather than to confirm a single established causal model. Partial Least Squares Structural Equation Modeling (PLS-SEM) was therefore selected because it is well-suited for prediction-oriented research involving complex models with multiple latent variables and moderating effects.

PLS-SEM was used to analyze both the measurement and structural models, including moderation effects from demographic variables such as age and sex. This design enabled the researcher to quantify associations among constructs and predict teachers’ behavioral intention to adopt generative AI tools in English language instruction, rather than actual usage behavior. PLS-SEM is particularly appropriate for models that integrate multiple theoretical frameworks and emphasize prediction and variance explanation in social science contexts (Cortez et al., 2024).

The study was conducted in Cebu, a province in the central Philippines that offers a heterogeneous higher education environment encompassing both urban and rural campuses. Cebu was selected because it hosts a mix of metropolitan institutions located in highly urbanized cities and satellite campuses situated in semi-urban and rural municipalities, allowing the study to capture variability in institutional resources, technological infrastructure, and instructional contexts relevant to behavioral intention in AI adoption.

Two state universities were included: State University 1 (SU1) and State University 2 (SU2), each operating multiple campuses across Cebu, providing both rural and urban backdrops. SU1, the largest technological university in Cebu, contributed the most respondents, while SU2 primarily focused on teacher education programs. In addition, five local colleges were included: three city colleges (CitC1, CitC2, and CitC3) and two community colleges (ComC1, ComC2). Collectively, these institutions represent state and local higher education sectors across diverse geographic and institutional settings, thereby strengthening the study’s capacity to examine behavioral intention toward generative AI adoption in both rural and urban English language teaching environments.

The study employed purposive sampling, selecting 488 higher education teachers from state universities and local colleges in Cebu. Inclusion criteria required participants to have taught English-related courses within the past two academic years and to report prior exposure to generative AI tools (e.g., ChatGPT, Grammarly, and Bard). English-related courses included grammar, composition, linguistics, literature, English for Specific Purposes (ESP), and interdisciplinary courses delivered in English, such as purposive communication and research writing.

The sample comprised both formally trained English instructors and educators from other disciplines who regularly handled English-medium instruction, reflecting the instructional realities of Philippine higher education institutions. Representation across institutions ensured participation from urban-based campuses with stronger digital infrastructure as well as rural and semi-urban campuses with more limited technological resources, thereby enhancing contextual representativeness (see Table 1).

Minimum sample size requirements were met using the rule of 10 cases per estimated parameter in SEM, and the power analysis further confirmed the adequacy of the sample (Kline, 2023). Only complete responses from participants with prior AI exposure were retained for analysis to ensure the reliability and validity of the findings.

The study adapted validated instruments to measure all constructs. The UTAUT section was based on An et al. (2023) and measured performance expectancy, effort expectancy, social influence, facilitating conditions, and behavioral intention.

Behavioral intention was operationally defined as teachers’ self-reported intention to use generative AI tools in future English language teaching activities, rather than current usage or instructional effectiveness.

EVT constructs, perceived attainment value, perceived intrinsic value, perceived utility value, and perceived cost, were adapted from Ka et al. (2023). Perceived knowledge of AI was measured using a six-item scale developed by Gado et al. (2021) that captures teachers’ subjective confidence and understanding of AI tools. Perceived privacy concerns were measured using Coopamootoo’s (2023) instrument, focusing on anticipated ethical and data-related risks. All items were rated on a 5-point Likert scale. The instruments underwent expert validation for clarity, content relevance, and theoretical alignment. Reliability and validity were assessed during data analysis using established SEM criteria.

Data were collected over a one-month period following approval from the institutional research ethics committee. Formal permission was obtained from participating institutions through transmittal letters outlining the study’s objectives, procedures, and ethical safeguards. A hybrid data collection approach was employed, combining Google Forms for online respondents and printed questionnaires for on-site administration at campuses with limited internet access.

All participants provided informed consent and were assured of voluntary participation, confidentiality, and anonymity. Pretesting was conducted with a small group of respondents to ensure clarity and appropriate survey length. Paper-based responses were digitized and merged with online data for consistency. Ethical protocols, including anonymization and secure data handling, were strictly followed throughout the data collection process.

Descriptive statistics were generated using SPSS version 20. Structural model analysis was conducted using SmartPLS 4.0. Measurement model evaluation included assessment of convergent validity through Average Variance Extracted (AVE ≥ 0.50) and discriminant validity using the Fornell–Larcker criterion and the Heterotrait–Monotrait (HTMT) ratio. Internal consistency was assessed using Cronbach’s alpha, composite reliability, and rho_C, all exceeding the recommended threshold of 0.70.

Structural relationships were tested using bootstrapping with 5,000 resamples. Moderation analysis was performed using the product-indicator approach in PLS-SEM. Age was modeled as a continuous moderator, and sex as a categorical moderator, with interaction terms created between each moderator and the corresponding predictor constructs.

Significance was evaluated using t-values ≥ 1.96 and p-values < 0.05. This analytical approach enabled robust testing of both direct and moderated relationships across UTAUT, EVT, and the extended constructs, ensuring theoretical precision and empirical rigor in explaining teachers’ behavioral intention to adopt generative AI.

This section reports the measurement model’s psychometric properties, including construct reliability and validity, to ensure an accurate representation of the theoretical domains. Three types of validity were assessed: discriminant validity, convergent validity, internal consistency reliability, and predictive validity assessment following the guidance of Hair et al. (2020).

This section evaluates the hypothesized relationships among latent variables using PLS-SEM with bootstrapping (5,000 subsamples). Significance was determined at t > 1.96 and p < 0.05. Twelve of fourteen hypotheses were supported.

Table 6 summarizes the structural model results, with 12 of 14 hypotheses supported through bootstrapping (5,000 subsamples). Statistically significant paths affirmed the model’s predictive validity, particularly among core constructs from UTAUT, EVT, and the extended variables.

Performance expectancy (PE) significantly predicted attainment value (H01) (β = 0.542), and effort expectancy (EE) influenced perceived utility value (UV) (H02) (β = 0.450), confirming that perceptions of usefulness and ease of use shape motivational appraisals. This is consistent with the work of Farhan et al. (2019), who emphasized that simplicity in digital tool design can directly enhance user adoption, especially in education systems with varying levels of technological proficiency.

Social influence (SI) significantly affected perceived attainment value (AV) (H03) (β = 0.395), perceived intrinsic value (IV) (H04) (β = 0.426), and perceived cost (PC) (H05) (β = 0.153), reinforcing the impact of peer and institutional support on both motivation and perceived barriers. These findings align with Zamiri and Esmaeili (2024), who found that peer and institutional encouragement are particularly influential in academic environments where shared norms and collaborative learning drive innovation. When teachers see colleagues using AI meaningfully or receive leadership endorsement, their sense of purpose and control increases. The implications are clear: professional learning communities, peer mentoring, and visible administrative support can go a long way in building confidence and motivation to adopt AI.

Facilitating conditions (FC) influenced perceived utility value (UV) (H06) (β = 0.247) and perceived cost (PC) (H07) (β = 0.151), emphasizing the role of institutional support. These results are echoed by Zhao and Zhao et al. (2024), who found that infrastructure, access to training, and consistent tech support are critical in sustaining teacher engagement with AI tools. Notably, the structural environment in which teachers operate affects how they calculate both the benefits and the effort required for technology use in rural settings, such as one of the environments in this study.

Perceived attainment value (AV) predicted perceived knowledge of AI (PKA) (H08) (β = 0.512), supporting Bandura’s (2015) concept of self-efficacy, where mastery expectations are shaped by perceived goal relevance, as cited in Bardach et al. (2019). However, perceived intrinsic value (IV) did not affect perceived knowledge of AI (PKA) (H09) (β = 0.022), indicating that enjoyment alone does not drive competence. This echoes the findings of Novak and Schwan (2021), who argue that while affective responses can foster curiosity, they must be paired with hands-on learning experiences to result in actual knowledge gains.

Perceived utility value (UV) positively influenced both perceived knowledge of AI (PKA) (H10) (β = 0.319) and perceived privacy concerns (PPC) (H11) (β = 0.375), suggesting that practical engagement increases both confidence and ethical awareness. The outcome aligns with recent insights by Durango et al. (2024), who note that higher engagement with AI increases users’ sensitivity to data-related risks, even as it fosters skill development.

Perceived cost (PC) also predicted perceived privacy concerns (PPC) (H12) (β = 0.095) indicating that teachers who anticipate a high investment of time or effort may also feel more cautious about data exposure (Beirat et al., 2025). These insights underscore the need for proactive digital ethics training that is closely tied to usability and accessibility.

The strongest path emerged between perceived knowledge of AI (PKA) and behavioral intention (BI) (H13) (β = 0.791), confirming that self-assessed competence is the primary driver of adoption. This finding aligns with studies by Kelly et al. (2023) and Kim and Lee (2022), which emphasize that knowledge, rather than general positivity, is a stronger predictor of AI engagement in educational settings.

Two hypotheses were not supported. First is perceived intrinsic value (IV) did not affect perceived knowledge of AI (PKA) (H09) (β = 0.022). In addition to what was stated above, this result suggests that curiosity alone does not guarantee understanding without training. This finding aligns with recent studies indicating that enjoyment and interest alone foster engagement but do not automatically lead to increased confidence or competence (Zhao et al., 2022; Lim and Wang, 2023). In other words, teachers are curious and enthusiastic about experimenting with generative AI, but without formal training or structured learning experiences, that enthusiasm does not translate into a solid sense of understanding. This highlights the need for institutions to go beyond sparking interest and offer sustained professional development programs that cultivate engagement and mastery.

Secondly, perceived privacy concerns (PPC) did not significantly influence behavioral intention (BI) (H14) (β = −0.033), because in resource-constrained contexts like the Philippines, perceived usefulness outweighs ethical reservations. This indicates that, in the present model, PPC does not explain additional variance in BI beyond that accounted for by other predictors, particularly perceived knowledge of AI (PKA) and the perceived value constructs. The nonsignificant effect should be interpreted as a model-based result rather than as evidence that privacy does not matter in practice; it suggests that privacy concerns may operate through alternative mechanisms (e.g., shaping institutional policy preferences or actual usage behavior) that were not directly tested in this study.”

These findings suggest that affective interest and ethical concerns, while relevant, do not directly translate into behavioral intention to adopt without supportive infrastructure and capacity-building. Institutions should therefore prioritize teacher training, recognize the value of alignment, and strengthen environmental supports while embedding privacy education into broader AI literacy programs. Together, the findings confirm that behavioral intention is driven by the interplay of cognitive, motivational, and contextual factors, requiring localized, multifaceted implementation strategies.

Throughout the structural model assessment, behavioral intention is treated as a cognitive–motivational outcome representing teachers’ future willingness to use generative AI, rather than as evidence of actual AI adoption or usage behavior.

This section enumerates the practical and pedagogical implications of the study’s findings. The implications enumerated from this study focus on factors shaping behavioral intention to use generative AI and do not assume actual adoption or implementation outcomes.

Drawing on the Unified Theory of Acceptance and Use of Technology (UTAUT), Expectancy-Value Theory (EVT), and extended constructs such as perceived knowledge of AI (PKA) and perceived privacy concerns (PPC), the study offers a nuanced understanding of how English language teachers in higher education evaluate and adopt generative AI. The discussion is organized under two thematic categories: enhancing pedagogical strategies and integrating technology into the curriculum.

First, the study confirms that performance expectancy (PE), perceived utility value (UV), and perceived knowledge of AI (PKA) significantly influence behavioral intention (BI). Teachers are more likely to adopt AI when they perceive it as pedagogically useful and feel confident in its use. AI tools like ChatGPT and Grammarly support personalized writing feedback and revision, especially in large classes (Zhou et al., 2023). Teachers can frame AI suggestions as springboards for discussion, promoting learner autonomy and metalinguistic awareness (García-Sánchez et al., 2023). Similarly, reflective tasks using AI-generated perspectives can foster critical thinking, provided students are guided to evaluate the credibility and tone of such content (Meihami and Meihami, 2023). The study shows that perceived intrinsic value (IV) alone does not predict perceived knowledge of AI (PKA), suggesting that enjoyment must be accompanied by structured reflection to yield meaningful learning. Teachers’ confidence also enables them to curate AI-generated output for differentiated instruction, especially in multilingual classrooms (Tuan, 2023; Belda-Medina and Goddard, 2024), while minimizing reliance on flawed or culturally misaligned content.

Second, perceived knowledge of AI (PKA) emerged as the strongest predictor of behavioral intention (BI), underscoring the need to institutionalize AI literacy within English language teaching programs. Curricula can include modules on AI-assisted summarization, vocabulary expansion, and lesson planning tailored to local linguistic contexts (Sandoval, 2024). Faculty development programs are essential to ensure that instructors are equipped to deliver these components effectively (Florendo and Alvarez, 2023). At the same time, assessment redesign is critical. Since AI can perform basic tasks like summarization and grammar correction, instructors should prioritize synthesis, critique, and ethical analysis to encourage higher-order thinking (Valbuena and Tongco, 2024). Perceived privacy concerns (PPC), although not statistically significant in predicting behavioral intention (BI), remain moderately present and warrant curriculum-embedded discussions on the ethical use of AI and data protection (Dizon and Espina, 2023).

Third, the study correspondingly finds that social influence (SI) positively affects perceived attainment value (AV) and perceived intrinsic value (IV), highlighting the importance of peer communities. Institutions can establish AI Teaching Circles to foster collaborative experimentation and informal mentoring, which build faculty confidence (Marquez and Austria, 2024). However, these communities require institutional recognition and resource support to be sustainable. Infrastructure disparities also pose a barrier. Teachers in rural or under-resourced campuses may lack access to reliable internet or AI subscriptions. In contexts such as Philippine higher education, institutions must audit technological needs and consider alternatives such as offline AI tools or localized content banks (Rivera and Mallari, 2024). Even incremental improvements in usability and support systems can significantly boost utility value.

Furthermore, this study found that even when teachers are aware of potential privacy or data risks associated with AI, these concerns do not significantly deter them from intending to use it. This result contrasts with earlier studies conducted in Western contexts, where privacy and surveillance issues are often cited as key barriers to AI adoption (Dwivedi et al., 2021; Mittelstadt, 2019). However, in developing or resource-constrained settings like the Philippines, the urgency to improve teaching tools and access to innovative technologies may outweigh abstract concerns about data security, reflecting a contextual prioritization of usefulness over risk among Cebu-based educators. This is similar to results in other third-world contexts, where perceived benefits often outweigh abstract concerns about data security (Ifinedo, 2012).

Lastly, consistent with findings from higher education studies in Arab contexts, where perceived usefulness and self-efficacy consistently outweigh privacy concerns in predicting intention (Sallam et al., 2025; Al-Emran et al., 2025; Salah and Ayyash, 2025), the present results suggest that teachers prioritize instructional value and competence development when forming intentions to use AI. These comparative findings indicate that across diverse cultural settings, including the Philippines and Arab higher education systems, motivational and cognitive factors exert stronger influence on behavioral intention than ethical risk perceptions at the intention-formation stage.

Overall, the study’s findings suggest that the effectiveness of AI integration in ELT is largely determined by perceived instructional benefits, confidence, and institutional support, rather than intrinsic interest or demographic variables. Teachers are willing to adopt AI when it is pedagogically relevant, ethically grounded, and institutionally supported. Ethical concerns, while secondary in behavioral prediction, should still be addressed through proactive training and policy safeguards. A triadic approach anchored in pedagogical design, digital ethics, and infrastructural support will help ensure that AI serves as a strategic partner in advancing language education outcomes in Philippine higher education.

Responding to gaps in the literature on generative AI adoption in English language teaching, this study examined the behavioral intention of higher education English language teachers in Cebu, Philippines, by integrating UTAUT, EVT, and extended constructs of perceived AI knowledge and perceived privacy concerns. The findings show that technology-related enablers influence behavioral intention indirectly through value-based motivational mechanisms. Performance expectancy predicts perceived attainment value; effort expectancy predicts perceived utility value; social influence shapes perceived attainment value, intrinsic value, and perceived cost; and facilitating conditions influence perceived utility value and perceived cost. Among downstream mechanisms, perceived attainment value and perceived utility value significantly predict perceived knowledge of AI, which emerges as the strongest direct predictor of behavioral intention. In contrast, perceived intrinsic value does not significantly predict perceived knowledge of AI, and perceived privacy concerns do not significantly predict behavioral intention, indicating that enjoyment and ethical concerns alone do not directly shape intention once perceived usefulness and self-assessed competence are considered.

With respect to demographic influences, age does not moderate any of the structural relationships, suggesting that teachers’ evaluations of generative AI are consistent across age groups in this context. Sex moderates only the relationship between effort expectancy and perceived utility value, while all other pathways remain invariant across genders. Taken together, these findings indicate that behavioral intention to adopt generative AI in English language teaching is primarily shaped by perceived usefulness, value alignment, and confidence, rather than by demographic characteristics. The study underscores the importance of institutional strategies that prioritize AI literacy, professional development, and supportive implementation environments to enable responsible and pedagogically aligned AI integration across diverse higher education contexts.

Several limitations should be acknowledged when interpreting the findings. First, the study employed a cross-sectional research design, which limits the ability to draw causal inferences or to examine changes in teachers’ perceptions and intentions over time. Behavioral intention was measured at a single point, and shifts in intention as teachers gain more experience with generative AI were not captured.

Second, all constructs were measured using self-reported data, which may be subject to social desirability bias and common method variance. In particular, perceived knowledge of AI reflects teachers’ subjective confidence rather than objectively assessed competence, and behavioral intention does not necessarily translate into actual classroom use.

Third, the study is geographically bounded to higher education institutions in Cebu, Philippines. While Cebu provides a diverse mix of urban and rural campuses, the findings may not be fully generalizable to other regions of the Philippines or to higher education systems with different policy environments, infrastructure conditions, or cultural norms. These limitations suggest the need for cautious interpretation beyond the studied context.

Future research may extend this work in several directions. Longitudinal studies are needed to examine how behavioral intention evolves into actual AI adoption and sustained use, particularly as institutional policies, training initiatives, and AI literacy programs mature. Such designs would allow researchers to test whether perceived knowledge of AI continues to predict usage behavior over time.

Comparative and cross-regional studies are also recommended to assess whether the integrated UTAUT–EVT framework operates similarly across different Philippine regions and in other higher education contexts, both developed and developing. Expanding the model to include objective measures of AI competence, classroom implementation data, or policy awareness may further strengthen explanatory power. Finally, future research could explore additional moderators, such as teaching experience, disciplinary background, or institutional AI governance maturity, to refine the understanding of how contextual conditions shape teachers’ engagement with generative AI.