Sage Journals: Discover world-class research

Abstract

While the Technology Acceptance Model (TAM) has been widely applied in AI in Education (AIEd) research, comprehensive evidence on TAM relationships and how cultural and non-cultural factors moderate them remains limited. To address this gap, this meta-analysis synthesizes 27 studies with over 12,000 participants. The results reveal that TAM relationships in AIEd exhibit large and significant positive correlations (r > .5), following similar ordering patterns (e.g., PU → BI > PEU → PU > PEU → BI) as other domains, while being stronger than those reported in other educational domains. In moderating core TAM relationships, cultural dimensions exhibited striking patterns with Uncertainty Avoidance showing opposite effects compared to Power Distance, Individualism, and Masculinity; education level demonstrated consistent effects, while technology type and income level showed pathway-specific effects. These findings validate TAM relationships in AIEd and provide evidence-based guidance for stakeholders implementing AI in diverse educational contexts.

Keywords

technology acceptance model artificial intelligence technology acceptance technology adoption education cultural dimensions

Introduction

The rapid integration of Artificial Intelligence (AI) technologies in educational settings has transformed the educational landscape (Zawacki-Richter et al., 2019). AI technologies are reshaping educational experiences and enhancing decision-making processes within educational institutions (Holmes et al., 2019). Recent meta-analyses indicate that AI technologies not only improve learning outcomes (Dai et al., 2024; Derakhshan et al., 2024; X. Wu & Li, 2024; R. Wu & Yu, 2024; Zheng et al., 2023) but also have a substantial impact on student engagement (Lo et al., 2024).

Despite these promising outcomes, the successful adoption of AIEd faces significant challenges (Mafara & Abdullahi, 2024). One critical concern highlighted by international organizations such as UNESCO is that the widespread use of AI technologies in education is closely connected to issues of equity and inclusion, as persistent disparities in access to digital resources and AI-driven tools may reinforce existing educational inequalities (UNESCO, 2021). These challenges indicate that understanding the factors influencing AI acceptance in education has become crucial for promoting its effective integration and ensuring that the benefits of AI are broadly accessible to learners and educators (Strzelecki & ElArabawy, 2024; Sun et al., 2024).

To investigate the factors influencing technology acceptance, researchers have employed multiple theoretical frameworks, including the Theory of Planned Behavior (TPB; Ajzen, 1991), the Technology Acceptance Model (TAM; Davis, 1985), and the Unified Theory of Acceptance and Use of Technology (UTAUT; Venkatesh et al., 2003), along with extensions such as TAM2 (Venkatesh & Davis, 2000), TAM3 (Venkatesh & Bala, 2008), and UTAUT2 (Venkatesh et al., 2012). Among these frameworks, TAM has established itself as a core model in educational technology research, with its central constructs—Perceived Usefulness (PU) and Perceived Ease of Use (PEU)—serving as foundational elements that have been integrated into most subsequent technology acceptance theories (Davis & Granić, 2024).

Prior research has shown that the explanatory power of TAM can vary across different technological and cultural contexts (Scherer et al., 2019), suggesting that contextual factors play a critical role in shaping technology acceptance. Although TAM has shown robustness in studies of technology adoption (Davis & Granić, 2024), the distinctive characteristics of AIEd warrant systematic examination of TAM’s applicability in this emerging domain. For example, AI’s adaptive learning capabilities enable real-time personalization, predictive analytics provide anticipatory support, and autonomous decision-making functions create novel user experiences beyond conventional educational technologies (Zawacki-Richter et al., 2019). These characteristics may alter the relative importance of core TAM constructs. Moreover, as generative AI technologies continue to reshape educational practices, questions have emerged regarding whether traditional technology acceptance frameworks remain sufficient for explaining AI adoption (Mogaji et al., 2024). This uncertainty, combined with the rapid proliferation of AIEd across diverse cultural and institutional contexts, creates urgent needs for systematic synthesis to validate TAM’s explanatory power and identify context-specific patterns in this domain.

Existing research evaluating TAM’s explanatory power can be broadly categorized into two types: studies assessing TAM’s explanatory power without specific context (Feng et al., 2021; Q. Ma & Liu, 2004; Schepers & Wetzels, 2007; K. Wu et al., 2011; Yousafzai et al., 2007a, 2007b) and studies focusing on specific contexts such as healthcare or education (Scherer et al., 2019). While these studies have enhanced our understanding of TAM’s explanatory power, a comprehensive review of the extant literature reveals two significant research gaps. The first research gap is the lack of a comprehensive meta-analysis in AIEd context. The first research gap is the lack of a comprehensive meta-analysis specifically focused on AIEd. This gap is critical for three reasons. First, AIEd’s distinctive technological characteristics—adaptive algorithms, predictive analytics, and autonomous decision-making—fundamentally differ from conventional educational technologies, potentially altering TAM dynamics and necessitating empirical validation of the model’s applicability. Second, the rapid expansion of AIEd across diverse cultural and institutional settings creates practical urgency for understanding whether TAM predictions generalize universally or require context-specific adaptations. Third, existing primary studies show substantial inconsistency across all core TAM relationships within AIEd. For example, Dehghani and Mashhadi, 2024) reported strong effects (PEU-BI: r = .73; PU-BI: r = .81; PEU-PU: r = .84), while Lai et al. (2023) found substantially weaker correlations for the same relationships (r = .14, r = .09, and r = .08, respectively). This dramatic variation underscores the necessity for meta-analytic synthesis to determine robust effect sizes and identify contextual moderators.

The second gap we identified is the inadequate examination of cultural and contextual moderators. While previous studies have demonstrated the moderating effects of Hofstede’s cultural dimensions (Hofstede, 1980) on technology adoption, the findings regarding these cultural moderating effects show considerable variation across different contexts. Moreover, there is a lack of meta-analytic research examining the moderating effects of crucial factors specific to AIEd, such as: AI technology types, user types in educational settings, and national income levels.

Addressing these research gaps is essential for several reasons. First, it will provide a more accurate assessment of TAM’s applicability and effectiveness in the specific context of AIEd. Second, employing a multidimensional cultural framework can capture more precise differences in AI educational tool acceptance across various cultural backgrounds, potentially guiding the development of more culturally adaptive AI educational systems.

This study aims to bridge these research gaps through a meta-analysis of TAM in the domain of AIEd. Specifically, we address two critical research questions (RQs):

RQ1: What are the overall effect sizes of the relationships among TAM constructs in the AIEd domain?

RQ2: How do moderating factors (Hofstede’s cultural dimensions, technology type, education level, and country income level) influence the relationships among TAM constructs in the AIEd domain?

This study is expected to make three distinct contributions beyond prior TAM meta-analyses that have mainly focused on general educational technologies or other domains (e.g., C. Liu et al., 2024; Scherer et al., 2019; Tao et al., 2020). First, it provides the first systematic synthesis specifically within the AIEd domain, filling a critical gap as AI increasingly transforms educational practices. Second, it advances cross-cultural analysis by decomposing geographic classifications into specific cultural dimensions (e.g., power distance, individualism, uncertainty avoidance) and country/region-level income factors, revealing more granular moderation patterns. Third, it examines whether TAM relationships are context-contingent, varying systematically across educational levels, user roles, economic contexts, and cultural dimensions. These contributions will inform both TAM theory refinement and culturally-adaptive AI implementation strategies in education.

Literature Review

Technology Adoption Models

Multiple frameworks have been proposed to explain technology adoption, including the Theory of Reasoned Action (TRA; Fishbein & Ajzen, 1977), the Theory of Planned Behavior (TPB; Ajzen, 1991), TAM (Davis, 1985), and the Innovation Diffusion Theory (IDT; Rogers, 1962). Building on these, Venkatesh et al. (2003) developed UTAUT, later extended into UTAUT2 (Venkatesh et al., 2012). Among them, TAM and UTAUT are the most widely applied in educational contexts (Granić, 2022). In the specific domain of AIEd, TAM has been identified as the most frequently used adoption model (Al-Momani & Ramayah, 2024). Similarly, in studies focusing on teachers’ adoption of AI, TAM has also emerged as the dominant framework (Xue, Ghazali, et al., 2025). While other frameworks such as Self-Determination Theory (SDT; Deci & Ryan, 1985), Expectation Confirmation Theory (ECT; Oliver & Linda, 1981), and Task-Technology Fit (TTF; Goodhue & Thompson, 1995) have also been applied in some studies, their usage remains relatively limited. This further underscores the prominence of TAM and UTAUT in AIEd research. UTAUT can be regarded as an extension of TAM, as its two central predictors—Performance Expectancy (PE) and Effort Expectancy (EE)—are conceptually rooted in TAM’s Perceived Usefulness (PU) and Perceived Ease of Use (PEU). Recent reviews have confirmed this close relationship, noting that hypotheses involving PE and EE are the most frequently tested in UTAUT-based studies of educational technology adoption. For example, Xue et al. (2024) found that hypotheses related to PE and EE were examined more often than those involving other constructs. This pattern was further reinforced in Xue, Ghazali, et al.’s (2025) latest review of UTAUT2 in AIEd contexts. Even in newer frameworks such as the Artificially Intelligent Device Use Acceptance (AIDUA) model (Gursoy et al., 2019), constructs equivalent to PU and PEU remain central. These findings highlight that although many adoption models have been developed, TAM remains the foundational framework that continues to inform and shape research on technology acceptance in educational contexts.

Technology Acceptance Model (TAM)

The Technology Acceptance Model (TAM), initially proposed by Davis (1989), has become a fundamental framework for understanding user acceptance of new technologies in various fields, including education. As AIEd continues to evolve rapidly, TAM has emerged as the most frequently used model for assessing user acceptance of AI technologies in educational contexts (Kelly et al., 2023).

TAM comprises five core constructs: Perceived Usefulness (PU), Perceived Ease of Use (PEU), Attitude (ATT), Behavioral Intention (BI), and Actual Use (AU). The model posits that AU is determined by BI, which is influenced by ATT, PU, and PEU. Additionally, PU and PEU affect ATT, while PEU influences PU (Davis, 1989). The extensive application of TAM has led to numerous literature reviews ( e.g., Al-Emran & Granić, 2021; Alomary & Woollard, 2015; Alshammari & Rosli, 2020; Chang et al., 2010; Chuttur, 2009; Doulani, 2019; Gupta et al., 2022; Lee et al., 2003; Legris et al., 2003; Marangunić & Granić, 2015; Mortenson & Vidgen, 2016; Silva, 2007; Turner et al., 2010; Yucel & Gulbahar, 2013) and meta-analyses (e.g., Feng et al., 2021; King & He, 2006; Ma & Liu, 2004; Marikyan et al., 2023; Schepers & Wetzels, 2007; Scherer et al., 2019; K. Wu et al., 2011; Yousafzai et al., 2007a, 2007b), significantly enhancing our understanding of technology acceptance across various domains.

TAM in Education

In the education sector, TAM has been applied to a wide range of topics, including technology adoption (Granić & Marangunić, 2019), online learning (Mustafa & Garcia, 2021), mobile learning (Al-Emran et al., 2018; C. Liu et al., 2024; Mugo et al., 2017), learning management systems (Cavus et al., 2022), higher education (Rosli et al., 2022), e-learning (Abdullah & Ward, 2016), and technology adoption by teachers (Scherer et al., 2019). Despite the varied themes, several consistent findings emerge regarding study subjects, sample regions, and research methodologies. Students are the most frequently surveyed subjects, higher education is the most examined level (Granić & Marangunić, 2019), Asian countries dominate sample regions (Al-Emran et al., 2018), and quantitative methods are predominantly used (Rosli et al., 2022). These findings deepen our understanding of TAM’s application in education.

TAM in AIEd

Despite TAM’s widespread use in AIEd, its robustness within this specific domain remains unclear. For instance, a recent systematic review of TAM in AI higher education contexts found that while some relationships such as PU-Attitude (AT), AT-Behavioral Intention (BI), and BI-Actual Use (AU) were consistently significant, others like PU-BI (80% significance) and PEU-BI (72% significance) were less stable across studies (Xue, Mahat, et al., 2025). However, this review relied primarily on significance counts and did not employ meta-analytic techniques, leaving open the question of whether TAM relationships are statistically robust in AIEd. The rapid development of AIEd introduces new variables and considerations that may affect TAM’s applicability and explanatory power. Previous meta-analyses have demonstrated the robustness of the TAM model in various contexts, such as technology adoption by teachers (Scherer et al., 2019) and mobile learning acceptance (C. Liu et al., 2024). Simultaneously, these meta-analyses reveal significant heterogeneity in the relationships between TAM variables across different studies. Despite TAM being the most commonly used technology acceptance model in the domain of AIEd, the robustness of TAM within this domain, the existence of heterogeneity, and the sources of such heterogeneity remain unclear. Previous meta-analyses have identified moderating effects of technology type (Schepers & Wetzels, 2007; Yousafzai et al., 2007b), education level (Abu-Shanab, 2011; C. Liu et al., 2024; Sabah, 2016; Tarhini et al., 2014), user type (Schepers & Wetzels, 2007), and culture (Jan et al., 2024; Y. Zhang et al., 2018) on TAM relationships. However, the existence and direction of these moderating effects within the specific context of AIEd remain unexplored, necessitating a focused meta-analysis of TAM in AIEd contexts.

Building on this gap, the present study examines five moderators: technology type, education level, user type, income level, and cultural dimensions. These moderators were selected because they capture critical contextual differences that may explain inconsistencies in TAM findings. Specifically, technology type distinguishes between general-purpose AI tools (e.g., ChatGPT), which are versatile and not tied to a fixed context, and domain-specific AI tools designed for particular educational tasks. User type differentiates students and teachers, whose perspectives on AI adoption are shaped by their distinct roles, with teachers focusing on teaching effectiveness and professional development and students prioritizing learning convenience and experience. Education level matters because acceptance patterns vary across K–12 and higher education contexts. Income level was chosen over coarse geographical groupings (e.g., Asia vs. non-Asia), as it offers a more nuanced understanding of socioeconomic disparities and aligns with UNESCO’s call to address educational inequalities in less-developed regions (UNESCO, 2021). Finally, cultural dimensions, which have been shown to influence both technology adoption (Jan et al., 2024; Y. Zhang et al., 2018) and educational practices (Ouyang et al., 2025), are introduced as moderators and will be elaborated in the following section.

Cultural Dimensions in TAM and AIEd

Hofstede (2011) defined culture as “the collective programming of the mind that distinguishes the members of one group or category of people from others” and proposed six main cultural dimensions: (a) Power Distance, (b) Uncertainty Avoidance, (c) Individualism/Collectivism, (d) Motivation Toward Achievement and Success, (e) Long/Short Term Orientation, and (f) Indulgence/Restraint. Power distance refers to the extent to which less powerful members accept and expect unequal power distribution. Uncertainty avoidance indicates a society’s tolerance for ambiguity and unstructured situations. Individualism/collectivism measures the degree of interdependence among members. Motivation toward achievement and success defines high scores (decisive) as valuing competition and achievement, while low scores (consensus-oriented) prioritize care and quality of life. Long/short term orientation focuses on future rewards versus past and present values. Indulgence stands for a society that allows relatively free gratification of basic and natural human desires related to enjoying life and having fun, while restraint refers to a society that controls gratification of needs through strict social norms.

Hofstede’s cultural dimensions have also been widely applied in educational contexts. A recent systematic review of 28 studies showed that cultural dimensions such as power distance, individualism–collectivism, and long-term orientation significantly influence educational management, leadership, and teaching–learning practices (Ouyang et al., 2025). These findings highlight the importance of cultural frameworks in understanding educational behaviors.

Recent studies have demonstrated the moderating effects of Hofstede’s cultural dimensions on technology adoption. For instance, Y. Zhang et al. (2018) found that these cultural dimensions significantly moderated electronic banking adoption. Similarly, a recent meta-analysis by Jan et al. (2024) confirmed the significant moderating effects of Hofstede’s cultural dimensions on technology adoption in general. However, it’s important to note that the specific moderating effects revealed in these studies showed some differences. These findings suggest that the moderating effects of Hofstede’s cultural dimensions may vary depending on the type of technology being adopted. This variation implies that the moderating effects of cultural dimensions on TAM relationships in the context of AIEd might differ from those observed in other technological domains. Therefore, there is a clear need to specifically examine how Hofstede’s cultural dimensions moderate TAM relationships within the AIEd context.

Examining the moderating role of cultural dimensions in TAM within AIEd is crucial as AIEd expands globally. Utilizing Hofstede’s multidimensional cultural framework can provide a nuanced understanding of how users from diverse cultural backgrounds accept AI educational tools. This analysis is essential for developing culturally adaptive AI education systems, enhancing their acceptability and effectiveness across various settings. By investigating how Hofstede’s cultural dimensions moderate TAM relationships in AIEd contexts, researchers can contribute to more effective, culturally sensitive AI educational technologies. This approach not only advances TAM theory in cross-cultural AIEd applications but also provides strategic guidance for the global promotion of AI educational technologies, opening new avenues for cross-cultural research in the AIEd field.

Methodology

Search Strategy

The literature search was conducted across four databases selected for their complementary strengths in capturing high-quality, peer-reviewed research: Scopus and Web of Science (comprehensive multidisciplinary coverage with rigorous indexing standards), ERIC (specialized education research), and IEEE Xplore (technology implementation studies). This combination was designed to ensure comprehensive retrieval of methodologically rigorous TAM-based studies while maintaining research quality standards appropriate for meta-analytic synthesis.

Our systematic search strategy incorporated terms and synonyms aligned with the focus concepts of the study, derived from prior systematic reviews on TAM (Granić & Marangunić, 2019), AI (Bond et al., 2024; Labadze et al., 2023), and educational themes (Labadze et al., 2023). The AI-related search terms were designed to be inclusive, encompassing both general descriptors (e.g., “artificial intelligence,”“intelligent systems”) and application-specific terms (e.g., “intelligent tutoring systems,”“chatbot”). This approach captures studies across different AI technologies and applications. Specific technical terms such as “deep learning” or “educational data mining” were not included as separate keywords because they represent underlying computational techniques rather than user-facing AI applications, and preliminary searches confirmed they retrieved predominantly technical methodology papers without yielding additional TAM-based studies. Compared with previous meta-analyses (e.g., Scherer et al., 2019; Vaccaro et al., 2024; Yu, 2023), this study adopted a more targeted query structure—refining TAM-related terms for conceptual precision while expanding AI-related descriptors to capture a broader range of educational applications. Searches were conducted on April 8, 2024, without publication date restrictions, ensuring thorough coverage. The specifics of these query strings are detailed in Table S1.

Criteria for Inclusion

The study’s inclusion criteria were applied in two phases. Initially, studies were screened based on language (only English publications), publication period (from 1986 to 2024), and source quality (peer-reviewed journal articles). The publication timeframe was set from 1986 to 2024. The starting point (1986) was chosen because the Technology Acceptance Model (TAM) was originally proposed in Davis’s doctoral dissertation, which marks the theoretical foundation of TAM-related research (Davis & Granić, 2024). The endpoint (2024) reflects the date when the literature search was conducted, ensuring comprehensive coverage of available studies up to the time of analysis. Furthermore, we restricted our review to peer-reviewed journal articles to ensure the inclusion of studies that have undergone rigorous quality control, thereby enhancing the reliability and validity of the meta-analytic findings. In the second phase, detailed screening involved checking for AI technologies or applications relevant to technology adoption, educational use of AI technologies or applications, utilization of TAM or its extended versions, empirical studies with hypothesis testing, and inclusion of core TAM variables: Perceived Usefulness (PU), Perceived Ease of Use (PEU), and Behavioral Intention (BI). Although constructs such as Performance Expectancy (PE) and Effort Expectancy (EE) in UTAUT are conceptually related to PU and PEU, we treated them separately in our analysis. This decision was made to maintain construct homogeneity and ensure comparability across studies, as PU and PEU are directly rooted in the original TAM framework and are operationalized more consistently across AIEd research. Exclusion criteria included studies that did not involve AI technologies, applied TAM in non-educational contexts, lacked empirical research or hypothesis testing, did not include PU, PEU, and BI, or were opinion pieces, editorials, or retracted articles.

Identification of Relevant Publications

During the initial database screening, 516 papers were identified: Web of Science (n = 167), Scopus (n = 282), ERIC (n = 34), and IEEE (n = 33). After removing 157 duplicates and excluding 4 book chapters, 355 publications remained. These were screened by titles and abstracts against the inclusion criteria, resulting in the exclusion of 278 papers. The remaining 77 papers underwent full-text screening, where 37 were excluded mainly due to the lack of TAM focus or absence of hypothesis testing. This rigorous process yielded 40 studies for quality appraisal.

Quality Appraisal

The 40 articles that satisfied all inclusion criteria underwent a meticulous re-reading and evaluation process to assess their quality. The quality appraisal process was conducted based on criteria developed by Claro et al. (2024) and Zhao et al. (2021), as presented in Table 1. Each criterion included a clear description to ensure a consistent understanding among the researchers involved.

Table 1.

Quality Criteria of the Selected Articles.

Quality criteria
Theoretical dimension:
1. Is the concept clearly defined?
2. Are the research objectives clearly specified?
3. Is the study designed to achieve the objectives
Methodological dimension:
4. Is the instrument clearly described and based on the design?
5. Is the instrument validated?
6. Is the instrument provided, and does it exhibit face validity?
7. Is the sample described, and is its size sufficient for the proposed analyses?
Findings dimension:
8. Are the research questions answered?
9. Are the conclusions clearly described and based on the results?

Each criterion had three possible responses: “Yes, complies” (1 point), “Does not comply” (0 points), and “Partially complies” (0.5 points). Articles needed a minimum score of 7 out of 9 to qualify. Two independent reviewers conducted the appraisal, and out of 360 coding decisions, 345 were consistent, yielding an observed agreement of 95.8%. Disagreements were resolved through discussion until consensus was reached.

As a result, 13 studies were excluded for failing to meet the quality threshold. The remaining 27 studies were deemed of sufficient quality to address the research questions. Individual study scores are provided in Table S2 for transparency.

Data extraction process is illustrated in a PRISMA flow diagram (Page et al., 2021) in Figure 1. The systematic screening process involved multiple stages: initial database searches identified 516 publications across four databases (Web of Science, Scopus, ERIC, and IEEE Xplore). After removing 157 duplicates and 4 no-peer-reviewed records, 355 publications remained for title and abstract screening. This screening phase excluded 278 papers that did not align with inclusion criteria, leaving 77 studies for full-text review. During full-text screening, 37 studies were further excluded primarily due to lack of TAM focus or absence of hypothesis testing, resulting in 40 studies advancing to quality appraisal. Finally, 13 studies were excluded for failing to meet the quality threshold (minimum score of 7 out of 9), yielding a final sample of 27 studies of sufficient quality for meta-analytic synthesis.

Figure 1.

PRISMA flow: Data extraction procedure.

Included Publications

The final pool of 27 selected publications for analysis is detailed in Table 2. Table 2 provides an overview of the studies that met all inclusion and quality criteria, forming the basis for our meta-analysis.

Table 2.

Characteristics of Included Publications.

Authors and year	Source of literature	Country/region	Education level	Subject	Technology	Income level
Al Darayseh (2023)	Computers and Education: Artificial Intelligence	UAE	K12	Teacher	AI applications	HI
Al Shamsi et al. (2022)	Education and Information Technologies	UAE	HE	student	AI-based voice assistants	LMI
Albayati (2024)	Computers and Education: Artificial Intelligence	South Korea	HE	student	ChatGPT	HI
Algerafi et al. (2023)	IEEE Access	China	HE	student	AI-based robots	UMI
Alrishan (2023)	Int J Learn Teach Educ Res	Oman, UAE	HE	pre-service teacher	ChatGPT	UMI
Awal and Haque (2024)	Journal of Applied Research in Higher Education	Bangladesh	HE	student	AI-Powered chatbot	HI
Ayanwale and Molefi (2024)	Int J Educ Technol High Educ	Lesotho	HE	student	chatbots	HI
Bilquise et al. (2024)	Education and Information Technologies	UAE	HE	student	academic advising chatbot	LMI
Choi et al. (2023)	International Journal of Human-Computer Interaction	South Korea	K12	Teacher	educational AI tools	HI
Dehghani and Mashhadi (2024)	Education and Information Technologies	Iran	CE	Teacher	ChatGPT	UMI
Esiyok et al. (2025)	International Journal of Human-Computer Interaction	Turkey	HE	student	AI-Powered chatbot	UMI
Huang et al. (2022)	Universal Access in the Information Society	China	K12	student	intelligent tutoring system	UMI
Gado et al. (2022)	Psychology Learning and Teaching	Germany	HE	student	AI	UMI
Lai et al. (2023)	Computers and Education: Artificial Intelligence	China	HE	student	ChatGPT	UMI
G. L. Liu et al. (2024)	Computer Assisted Language Learning	China	CE & HE	learner	GPT	UMI
G. Liu and Ma (2024)	Innovation in Language Learning and Teaching	China	CE & HE	learner	GPT	HI
Y.-L. E. Liu and Huang (2025)	International Journal of Human-Computer Interaction	Taiwan	HE	student	AI-Based Text-to-Image Technology	HI
Ma and Lei (2024)	Asia Pacific Journal of Education	China	HE	pre-service teacher	AI	LMI
Masa’deh et al. (2024)	International Journal of Data and Network Science (most)	Jordan	HE	student	ChatGPT	LMI
Ni and Cheung (2023)	Education and Information Technologies	China	K12	student	AI-powered intelligent tutoring system	HI
Pillai et al. (2024)	Information Technology and People	India	HE	Teacher	AI-based teacher-bots	UMI
Rahman et al. (2023)	Australasian Journal of Educational Technology	Bangladesh	HE	student	ChatGPT	HI
Sukkeewan et al. (2024)	International Journal of Information and Education Technology	Thailand	HE	student	smart learning platform	LMI
Tiwari et al. (2023)	Interactive Technology and Smart Education	Oman	K12 & HE	student	ChatGPT	UMI
C. Zhang et al. (2023)	Int J Educ Technol High Educ	Germany	HE	pre-service teacher	AI-Based Applications	UMI
B. Zou et al. (2025)	Computer Assisted Language Learning	China	HE	student	AI speech evaluation program	UMI
M. Zou and Huang (2023)	Frontiers in Psychology	China	HE	student	ChatGPT	HI

Note. K-12 = primary and secondary education; HE = Higher Education; CE = Continuing Education; UAE = United Arab Emirates; Int J Educ Technol High Educ = International Journal of Educational Technology in Higher Education; Int J Learn Teach Educ Res = International Journal of Learning, Teaching, and Educational Research; HI = High income; UMI = Upper middle income; LMI = Lower middle income.

Coding Process of Included Articles

A coding scheme was developed to categorize AI technology as “Vertical” (specific educational uses like tutoring systems) or “Horizontal” (general-purpose AI like ChatGPT). Education levels were classified as “K-12” or “Higher Level Education.” Income levels followed the World Bank’s 2021 criteria (World Bank, 2022). Cultural dimensions were coded using Hofstede’s model (Hofstede, 1980), with data from Hofstede Insights (2024). To conduct moderator analyses, each dimension was dichotomized into higher versus lower level groups. The cutoff points were determined using the median value of each dimension across all countries represented in the included studies. This median-split approach ensured balanced subgroup sizes and maximized statistical power. Of the 27 studies included in the meta-analysis, 24 were eligible for cultural moderation analysis because three studies (Alrishan, 2023; Ayanwale & Molefi, 2024; Tiwari et al., 2023) originated from countries not covered by Hofstede’s database (e.g., Lesotho, Oman) or contained mixed-country samples, preventing reliable assignment of cultural dimension scores. The resulting median-based cutoffs were as follows: power distance = 77 (k = 12 higher level, k = 12 lower level), individualism = 43 (k = 14 higher level, k = 10 lower level), uncertainty avoidance = 60 (k = 13 higher level, k = 11 lower level), masculinity (motivation toward achievement and success)= 56 (k = 12 higher level, k = 12 lower level), long-term orientation = 77 (k = 12 higher level, k = 12 lower level), and indulgence = 26 (k = 14 higher level, k = 10 lower level), where k refers to the number of studies included in each subgroup. Other details included author, year, source, location, participant identity, count, and AI type. AI technologies were categorized as either horizontal or vertical applications based on their degree of cross-domain generality. Technologies with cross-domain generality were classified as horizontal AI (e.g., ChatGPT, AI-based chatbots, voice assistants), whereas those designed for specific domains were classified as vertical AI—in this study, referring to AI systems specifically developed for educational purposes or contexts (e.g., intelligent tutoring systems, teacher-bots, academic advising tools, AI-based assessment programs). This classification ensures transparency and consistency in the moderator coding process. Hypothesis testing outcomes (standardized path coefficients, Pearson correlation coefficients, and p-values) were recorded.

Statistical Approaches

The meta-analysis adhered to the data analysis methods employed by Tao et al. (2020). Random-effects models were used to pool effect sizes due to observed heterogeneity. The correlation coefficient (r) was used as the effect size, with regression coefficients (β) converted to r using Peterson and Brown (2005) formula. This conversion was applied to 2 of 27 studies (7.4%), with all β values (range: .137–.709) within the reliable range where conversion bias is minimal (Peterson & Brown, 2005). Each r coefficient was converted into its corresponding Fisher’s Z score for data analysis, and the synthesized Fisher’s Z score was transformed back to r to display the results (Lei et al., 2018). Confidence intervals (95% CI) were calculated with interpretations of .1 ≤ r < .3 as small, .3 ≤ r < .5 as medium, and r ≥ .5 as strong effects (Cohen, 1988). Heterogeneity was assessed using the I² statistic and Cochran’s Q test. I² values of 25%, 50%, and 75% indicated low, moderate, and high heterogeneity, respectively. A significant Q statistic suggested rejection of the null hypothesis of homogeneity (Higgins & Thompson, 2002). Egger’s regression test was used to examine publication bias, with p < .05 indicating potential bias (Egger et al., 1997). To address variability, subgroup analyses identified factors moderating TAM relationships. Between-group effects were assessed using the between-group heterogeneity coefficient (Q_B), and group homogeneity was tested as outlined by (Borenstein et al., 2021). Sensitivity analyses were conducted to identify potential outliers using standardized residuals (|z| > 2) and Cook’s distance (threshold: 4/k), following established procedures for meta-analytic outlier and influence diagnostics (Viechtbauer & Cheung, 2010). Statistical analyses were performed using the Metafor package (Viechtbauer, 2010) in R (R Core Team, 2023), including random-effects meta-analyses, heterogeneity assessments, subgroup analyses, sensitivity analyses (outlier detection and influence diagnostics), and publication bias tests (Egger’s regression). Forest plots and funnel plots were generated using MetaEssentials 1.5 (Suurmond et al., 2017).

To maintain statistical independence across studies, we ensured that each meta-analytic model included only one effect size per study for each theoretical relationship (e.g., PEU-PU, PU-BI, PEU-BI). No study contributed more than one effect size to the same relationship. Therefore, issues of dependent effect sizes within studies did not arise in this meta-analysis.

All moderator analyses employed categorical subgroup comparisons rather than continuous meta-regression. This approach was adopted for several reasons. First, some moderators of theoretical interest (e.g., technology type and user type) are inherently categorical, reflecting qualitative differences not adequately captured by continuous scales. Second, categorical analysis aligns with the predominant analytical practice in prior TAM meta-analyses (e.g., Tao et al., 2020), facilitating direct comparison with established findings. Third, to ensure methodological consistency and interpretability across all moderator analyses, we applied the same categorical framework to cultural dimensions, despite their continuous nature in Hofstede’s model. This unified approach provides context-specific effect sizes that are directly comparable across moderator categories and more interpretable for educational practitioners and policymakers.

Results

Meta-Analysis of the Original TAM Relationships in the AIEd Domain

The meta-analysis examined seven correlations among five TAM constructs: PEU, PU, AT, BI, and AU. The results, along with tests for heterogeneity and publication bias, are presented in Table 3. Forest plots for all seven TAM relationships are presented in Figures 2 to 8. In each plot, individual studies are shown as blue dots (dot size indicates study weight) with horizontal lines representing 95% confidence intervals. The bottom row displays the meta-analytic summary: the green dot represents the combined effect size, the black interval shows the 95% confidence interval for the combined effect, and the wider green interval shows the 95% prediction interval (Hak et al., 2016). All analyses used random-effects models. Detailed statistics are provided in Table 3.

Table 3.

Meta-Analytic Results for Pairwise Relationships.

Correlations	k	Total sample size	Effect size [95% CI]	Homogeneity (Q-value)	I² (%) [95% CI]	Egger’s test (p-value)
PEU-PU	27	12,124	0.63 [0.56, 0.69]	723.58***	96.96 [96.33, 96.47]	−0.35 (.73)
PEU-BI	27	12,124	0.56 [0.49, 0.63]	740.53***	96.65 [96.41, 96.55]	0.76 (.45)
PU-BI	27	12,124	0.69 [0.62, 0.75]	1380.23***	97.85 [98.09, 98.13]	0.57 (.58)
PEU-AT	9	4,400	0.62 [0.52, 0.71]	169.31***	95.95 [94.99, 95.43]	0.10 (.92)
PU-AT	9	4,400	0.68 [0.57, 0.76]	208.93***	97.02 [95.99, 96.27]	0.13 (.90)
AT-BI	9	4,400	0.72 [0.61, 0.80]	271.24***	97.36 [96.94, 97.11]	0.42 (.69)
BI-AU	5	3,366	0.61 [0.37, 0.77]	441.96***	98.76 [99.08, 99.10]	1.19 (.32)

Note. Effect size [95% CI]: pooled correlation coefficient and its 95% confidence interval; Homogeneity (Q-value): heterogeneity test statistic with p-value in parentheses; I² [95% CI]: heterogeneity index and its 95% confidence interval; Egger’s test values are Z statistics. PEU = Perceived Ease of Use; PU = Perceived Usefulness; AT = Attitude; BI = Behavioral Intention; AU = Actual Use; k = number of studies.

***

p < .001.

Figure 2.

Forest plot for the PEU-PU relationship.

Figure 3.

Forest plot for the PEU-BI relationship.

Figure 4.

Forest plot for the PU-BI relationship.

Figure 5.

Forest plot for the PEU-AT relationship.

Figure 6.

Forest plot for the PU-AT relationship.

Figure 7.

Forest plot for the AT-BI relationship.

Figure 8.

Forest plot for the BI-AU relationship.

High levels of heterogeneity were detected across all correlations (I² ranging from 95.95% [95% CI: 94.99%, 95.43%] to 98.76% [95% CI: 99.08%, 99.10%]), with narrow confidence intervals indicating stable and reliable heterogeneity estimates. Egger’s regression tests indicated no evidence of publication bias across all relationships (all p > .05). Funnel plot asymmetry was further examined through visual inspection (Figures S1–S7), which revealed no substantial asymmetry across any of the seven relationships. As illustrated in the forest plots (Figures 2 –8), nearly all individual studies showed positive correlations, with consistent directionality across the field despite considerable variation in effect magnitude. The meta-analysis results showed that all pooled correlations were positive and significant. The 95% confidence intervals for the correlations between PU and behavioral intention, PEU and PU, and PEU and behavioral intention were narrower than those for other correlations, indicating robustness and consistency across studies. All pooled relationships showed large effect sizes (r > .5), with significant positive correlations among the five TAM constructs. Specifically, AT (r = .72) and PU (r = .69) exhibited the strongest correlations with BI. This was followed by PEU (r = .56). Both PU (r = .68) and PEU (r = .62) had strong correlations with AT. Additionally, behavioral intention (r = .61) showed a strong correlation with actual usage behavior. PEU also showed a strong correlation with PU (r = .63).

Moderators of the Original TAM Relationships in the AIEd Domain

The meta-analysis revealed significant heterogeneity in effect sizes across studies, with I² values ranging from 95.95% to 98.76%, indicating substantial variation (Higgins & Thompson, 2002). Therefore, a moderator analysis was conducted to examine the underlying causes of these variations. The analysis focused on three relationships: PEU with PU, PU with BI, and PEU with BI.

Detailed subgroup results are provided in Table S3, which reports the meta-analytic effect sizes and between-group heterogeneity statistics for each moderator. Overall, multiple moderators significantly influenced these relationships. Specifically, Technology Type showed that Vertical Technology had stronger correlations than Horizontal Technology for PEU with PU (QB = 106.51, p < .001) and PU with BI (QB = 42.11, p < .001). Education Level showed that K12 education had stronger correlations than higher education for all three relationships (PEU with PU, PEU with BI, PU with BI: QB = 22.55, p < .001; QB = 31.49, p < .001; QB = 47.03, p < .001 respectively). User Type revealed that students had stronger correlations than teachers for PEU with PU (QB = 81.89, p < .001). Income Level indicated that lower income had stronger correlations for PEU with PU (QB = 63.03, p < .001), while higher income had stronger correlations for PU with BI (QB = 121.18, p < .001). Power Distance showed that lower power distance (<77) resulted in stronger correlations for all three relationships (QB = 13.25, p < .001; QB = 31.39, p < .001; QB = 128.79, p < .001). Individualism showed that lower individualism (<43) led to stronger correlations for all three relationships (QB = 92.18, p < .001; QB = 42.05, p < .001; QB = 3.95, p < .05). Motivation revealed that lower motivation levels (<56) produced stronger correlations for all three relationships (QB = 86.92, p < .001; QB = 147.49, p < .001; QB = 250.37, p < .001). Uncertainty Avoidance showed that higher uncertainty avoidance (<60) had stronger correlations for all three relationships (QB = 8.46, p < .01; QB = 60.02, p < .001; QB = 144.32, p < .001). Long-Term Orientation indicated that lower long-term orientation (<77) had stronger correlations for PEU with PU (QB = 51.49, p < .001). Indulgence revealed that higher indulgence (≥26) was associated with stronger correlations for PEU with PU (QB = 31.26, p < .001), while lower indulgence (<26) resulted in stronger correlations for PEU with BI (QB = 27.73, p < .001).

As summarized in Table 4, moderating effects varied substantially across relationships. Education Level was the only non-cultural moderator demonstrating significant effects across all three relationships, with K-12 education consistently showing stronger correlations than higher education. Among the six cultural dimensions examined, four—Power Distance, Individualism, Masculinity, and Uncertainty Avoidance—showed significant effects across all three relationships.

Table 4.

Summary of Moderator Effects and Directions Across Core TAM Relationships.

Moderators	PEU-PU		PEU-BI		PU-BI
Moderators	Q_B	Direction	Q_B	Direction	Q_B	Direction
Technology type	106.51***	Ho < V	0.60	—	42.11***	Ho < V
Education level	22.55***	He < K	31.49***	He < K	47.03***	He < K
User type	0.27	—	81.89***	T < S	129.19***	T < S
Income level	63.03***	L > Hi	0.31	—	121.18***	L < Hi
Power distance	13.25***	L > Hi	31.39***	L > Hi	128.79***	L > Hi
Individualism	92.18***	L > Hi	42.05***	L > Hi	3.95*	L > Hi
Masculinity	86.92***	L > Hi	147.49***	L > Hi	250.37***	L > Hi
Uncertainty avoidance	8.46**	L < Hi	60.02***	L < Hi	144.32***	L < Hi
Long term orientation	51.49***	L > Hi	0.00	—	2.40	—
Indulgence	31.26***	L < Hi	27.73***	L > Hi	2.78	—

Note. “Direction” indicates which group shows a stronger relationship; “>” means the first group has a stronger effect; “<” means weaker; “—” denotes a non-significant moderator effect (p > .05). PEU = Perceived Ease of Use; PU = Perceived Usefulness; BI = Behavioral Intention; Ho = Horizontal technology; V = Vertical technology; He = Higher education; K = K–12 education; T = Teacher; S = Student; L = Low level (e.g., low income or low cultural index); Hi = High level (e.g., high income or high cultural index); Q_B = between-group heterogeneity coefficient.

p < .05. **p < .01. ***p < .001.

A critical pattern emerged in the direction of cultural moderating effects. Power Distance, Individualism, and Masculinity showed a consistent pattern: lower levels of these dimensions were associated with stronger relationships across all three pathways. In contrast, Uncertainty Avoidance showed the opposite directional pattern, with higher levels associated with stronger relationships across all three pathways.

Several moderators demonstrated reversal patterns across different relationships. Income level showed opposing effects across pathways, with lower-income contexts strengthening PEU-PU and higher-income contexts strengthening PU-BI. Indulgence similarly revealed directional reversal, with lower indulgence strengthening PEU-BI and higher indulgence strengthening PEU-PU.

Technology Type and User Type showed selective moderating effects on specific relationships rather than universal effects. Long-Term Orientation and Indulgence showed significant effects for only one or two of the three relationships, indicating more circumscribed influences compared to other cultural dimensions.

Sensitivity Analysis

To assess the robustness of our findings and address potential concerns about outliers, we conducted sensitivity analyses for all hypothesized relationships (Table 5). Outliers were identified using standardized residuals (threshold: |z| > 2) and Cook’s distance (threshold: 4/k). For relationships where outliers were detected, we re-estimated the pooled effect sizes after removing these studies. As shown in Table 5, outlier detection identified four studies as potential outliers across four relationships. Sensitivity analyses revealed that removing these outliers resulted in minimal changes to the pooled effect sizes (Δr ranging from + .015 to + .033), with all changes below the threshold of .05. These findings indicate that our results are robust and not driven by individual studies. The correlations between PEU-BI, AT-BI, and BI-AU showed no outliers, further supporting the stability of these relationships.

Table 5.

Sensitivity Analysis: Outlier Detection and Influence on Pooled Effect Sizes.

Correlations	k	Original r [95% CI]	Outliers detected	k After removal	r After removal [95% CI]	Δr	Interpretation
PEU-PU	27	.63 [0.56, 0.69]	Lai et al. (2023)	26	.64 [0.58, 0.70]	+.015	Robust
PEU-BI	27	.564 [0.49, 0.63]	None	—	—	—	Robust
PU-BI	27	.69 [0.62, 0.75]	Lai et al. (2023)	26	.70 [0.65, 0.75]	+.015	Robust
PEU-AT	9	.62 [0.2, 0.71]	Gado et al. (2022)	8	.65 [0.58, 0.72]	+.032	Robust
PU-AT	9	.68 [0.57, 0.76]	Gado et al. (2022)	8	.71 [0.64, 0.77]	+.033	Robust
AT-BI	9	.72 [0.61, 0.80]	None	—	—	—	Robust
BI-AU	5	.61 [0.37, 0.77]	None	—	—	—	Robust

Note. Outliers were identified using standardized residuals (|z| > 2) and Cook’s distance (4/k); “Robust” indicates that the removal of outliers had minimal impact (|Δr| < .05) on the pooled effect size. k = number of studies; r = pooled correlation coefficient; CI = confidence interval; Δr = change in pooled effect size after removing outliers. PEU = Perceived Ease of Use; PU = Perceived Usefulness; AT = Attitude; BI = Behavioral Intention; AU = Actual Use.

Discussion

This study evaluated AI acceptance in education using TAM frameworks through a meta-analysis of 27 studies with 12,124 participants. The findings indicate that the relationships among TAM constructs are consistently supported within the AIEd domain. Furthermore, relationships between PU, PEU, and BI are moderated by Hofstede’s cultural dimensions, technology type, education level, and country income level.

While previous meta-analyses have examined technology acceptance across various contexts (see Table 6 for detailed comparisons), this study represents the first meta-analytic synthesis that specifically focuses on TAM within the domain of AIEd.

Table 6.

Comparison of Results From Our Meta-Analysis With That From Previous Meta-Analysis Studies on Technology Acceptance.

Study	Technology type	ES [95% CI]
Study	Technology type	PEU-PU	PEU-BI	PU-BI
King and He (2006)	All types	0.49 [0.44, 0.54]	0.43 [0.37, 0.48]	0.59 [0.55, 0.63]
Schepers and Wetzels (2007)	All types	0.55 [0.50, 0.60]	0.47 [0.42, 0.52]	0.63 [0.58, 0.68]
Yousafzai et al. (2007b)	All types	0.44 [0.43, 0.55]	0.34 [0.32, 0.36]	0.55 [0.54, 0.56]
K. Wu et al. (2011)	All types	0.71 [0.70, 0.72]	0.58 [0.57, 0.59]	0.81 [0.80, 0.82]
Scherer et al. (2019)	All types in education	0.48 [0.45, 0.51]	0.42 [0.39, 0.46]	0.55 [0.52, 0.59]
Tao et al. (2020)	CHITs	0.65 [0.57, 0.72]	0.58 [0.49, 0.65]	0.66 [0.59, 0.71]
C. Liu et al. (2024)	mobile learning	0.57[0.52, 0.61]	0.51 [0.47, 0.55]	0.58 [0.54, 0.62]
The present study	AI in education	0.63 [0.56, 0.69]	0.56 [0.49, 0.63]	0.69 [0.62, 0.74]
		ATT-BI	PEU-ATT	PU-ATT	BI-AU
Schepers and Wetzels (2007)	All types	0.55 [0.41, 0.69]	0.54 [0.41, 0.67]	0.67 [0.57, 0.77]	0.54 [0.39, 0.68]
Yousafzai et al. (2007b)	All types	0.56 [0.54, 0.58]	0.45 [0.43, 0.47]	0.53 [0.51, 0.55]	0.46 [0.44, 0.49]
K. Wu et al. (2011)	All types	0.86 [0.85, 0.88]	0.67 [0.65, 0.68]	0.93 [0.92, 0.95]
Scherer et al. (2019)	All types in education	0.52 [0.48, 0.57]	0.53 [0.49, 0.56]	0.59 [0.55, 0.62]	0.46 [0.39, 0.54]
Tao et al. (2020)	CHITs	0.67 [0.51, 0.79]	0.69 [0.53, 0.80]	0.76 [0.63, 0.85]	0.71 [0.33, 0.89]
C. Liu et al. (2024)	Mobile learning	0.57 [0.50, 0.63]	0.52 [0.45, 0.58]	0.59 [0.51, 0.65]
The present study	AI in education	0.72 [0.61, 0.80]	0.62 [0.52, 0.71]	0.68 [0.57, 0.76]	0.61 [0.37, 0.77]

Note. PEU = Perceived Ease of Use; PU = Perceived Usefulness; ATT = Attitude; BI = Behavioral Intention; AU = Actual Usage Behavior.

The findings reveal that all synthesized relationships are positive and statistically significant, with notable variations in their strengths. Across the three core TAM relationships, PU-BI is stronger than PEU-PU, which in turn is stronger than PEU-BI. These patterns are consistent with previous meta-analyses across general contexts (King & He, 2006; Schepers & Wetzels, 2007; Wu et al., 2011; Yousafzai et al., 2007b), healthcare (Tao et al., 2020), and educational contexts (C. Liu et al., 2024; Scherer et al., 2019). Similarly, among relationships involving attitude, AT-BI is stronger than PEU-BI, and PU-ATT is stronger than PEU-ATT. Notably, the PU-BI and PU-ATT relationships demonstrate comparable strengths. These findings are consistent with the same meta-analytic literature noted above. Collectively, these patterns indicate that TAM constructs exhibit robust and consistent relationship patterns across various domains, including AIEd. A critical finding is that the strength of relationships among TAM constructs in AIEd exceeds those reported in previous meta-analyses of teacher technology acceptance (Scherer et al., 2019) and mobile learning (C. Liu et al., 2024). This disparity may be attributed to the unique characteristics of the AIEd domain, particularly its emphasis on personalized learning experiences. Enhanced personalization can strengthen user perceptions of usefulness and ease of use, thereby amplifying their influence on behavioral intention. These stronger relationships suggest that within AIEd contexts, PU and PEU exert particularly significant effects on BI.

To further understand the heterogeneity observed in these relationships, this meta-analysis examined both cultural and non-cultural contextual factors that moderate TAM relationships in AIEd.

This study reveals several factors moderating TAM relationships. Firstly, technology type moderates TAM constructs. Specifically, stronger connections were observed between PU-BI and PEU-PU in vertical AI technologies compared to horizontal ones in education. This aligns with findings by Schepers and Wetzels (2007) and Yousafzai et al. (2007b) regarding PU-BI but contrasts with Schepers and Wetzels (2007) on PEU-PU. Secondly, education level moderates TAM relationships, supporting previous studies (Abu-Shanab, 2011; C. Liu et al., 2024; Sabah, 2016; Tarhini et al., 2014). Stronger connections between PU, PEU, and BI were found in K12 settings compared to higher education. Thirdly, user type moderates TAM relationships, with students showing stronger correlations between PEU-BI than teachers, consistent with Schepers and Wetzels (2007). Fourthly, income level moderates TAM relationships. Stronger connections between PEU-PU were observed in lower-income countries/regions, while PU-BI connections were stronger in higher-income countries/regions.

Table 7 presents the moderating effects of Hofstede’s cultural dimensions on the relationships between PU, PEU, and BI, and compares these results with prior studies. The analysis revealed a complex pattern of cultural moderation in AIEd, characterized by both cross-cutting mechanisms and context-specific divergences from previous research.

Table 7.

Comparison of Results From Our Hofstede’s Cultural Dimensions Moderation Analysis With Previous Studies.

Study	Technology type	Hofstede’s cultural dimensions
Study	Technology type	Power distance	Individualism	Masculinity	Uncertainty avoidance	Long-term orientation	Indulgence
Jan et al. (2024)	All types	PEU-BI (O)PU-BI (−)	PEU-BI (+)PU-BI (+)	PEU-BI (−)PU-BI (+)	PEU-BI (−)PU-BI (−)	PEU-BI (+)PU-BI (+)
Y. Zhang et al. (2018)	Electronic banking	PEU-BI (O)PU-BI (−)	PEU-BI (+)PU-BI (O)	PEU-BI (O)PU-BI (−)	PEU-BI (O)PU-BI (+)	PEU-BI (−)PU-BI (−)
The present study	AIEd	PEU-PU (−)PEU-BI (−)PU-BI (−)	PEU-PU (−)PEU-BI (−)PU-BI (−)	PEU-PU (−)PEU-BI (−)PU-BI (−)	PEU-PU (+)PEU-BI (+)PU-BI (+)	PEU-PU (−)PEU-BI (O)PU-BI (O)	PEU-PU (+)PEU-BI (−)PU-BI (O)

Note. (+) indicates a significant positive moderation; (−) indicates a significant negative moderation; (O) indicates no significant moderation.

First, Uncertainty Avoidance showed an inverse directionality compared with Power Distance, Individualism, and Masculinity. While lower levels of the latter three dimensions strengthened all three TAM relationships, higher levels of Uncertainty Avoidance reinforced them (Table 7). In cultures characterized by high Uncertainty Avoidance according to Hofstede’s framework (Hofstede, 2011), individuals are theoretically posited to perceive novel and ambiguous situations with heightened concern. Applied to the AIEd context, the algorithmic opacity and pedagogical uncertainties inherent in AI systems (Zawacki-Richter et al., 2019) may activate this disposition. We propose that under these conditions, users intensify their reliance on assessments of PU and PEU as mechanisms for risk evaluation and mitigation. This interpretation is grounded in two observations: First, learners in high-uncertainty-avoidance cultures would engage in more deliberate evaluations of whether the technology genuinely delivers promised educational benefits and whether they can effectively operate it, thereby strengthening the association between these perceptions and behavioral intention. Second, conversely, in low-uncertainty-avoidance cultures, users may adopt AI tools with less intensive scrutiny of usefulness or usability, potentially weakening these relationships.

Interestingly, the moderating effect of Individualism observed in this study contrasts with previous meta-analytic findings (Jan et al., 2024; Y. Zhang et al. 2018), where Individualism typically strengthened the relationships between PU, PEU, and BI. One possible explanation for this divergence is the distinctly social and institutional nature of AI adoption in formal educational settings. We theorize that unlike technology adoption in many other domains, where decisions tend to be individually driven and motivated by personal efficiency or convenience, AIEd technologies are typically introduced within collective, norm-regulated educational contexts, where adoption decisions are shaped by the social influence and structural support of educational institutions, rather than by purely individual volition (Xue, Ghazali, et al., 2025). If this interpretation is correct, in collectivist cultures, PU and PEU may exert stronger effects on BI through conformity and social validation, whereas in individualistic cultures, autonomous and self-determined learning preferences may weaken these associations. This theoretical explanation, while consistent with cultural psychology literature, warrants empirical validation.

Consistent with previous studies (Jan et al., 2024; Y. Zhang et al., 2018), Power Distance negatively moderates the PU–BI relationship. However, unlike these studies, the present AIEd analysis also reveals significant negative moderation on the PEU–BI pathway (Table 7). This divergence on the PEU–BI pathway may reflect AI’s distinctive technical characteristics. Unlike conventional educational technologies with transparent interfaces, AI systems are characterized by algorithmic opacity and adaptive behaviors that users cannot fully predict or control (Zawacki-Richter et al., 2019). When learners interact with such opaque systems, perceived ease of use may become disconnected from actual system effectiveness. In high–Power Distance cultures, where individuals have limited autonomous decision-making authority (Hofstede, 1980), this technological opacity could further weaken the link between ease-of-use perceptions and behavioral intentions. Conversely, in domains such as electronic banking or general ICT use, where system functions are more transparent and predictable, perceived ease of use more reliably predicts adoption. In addition to these technical explanations, AIEd’s institutionalized adoption structures may further amplify Power Distance effects by constraining individual agency in technology-related decisions (Daniels & Greguras, 2014). Thus, Power Distance’s stronger moderation of the PEU–BI pathway in AIEd likely reflects an interplay between AI’s inherent opacity and the hierarchical dynamics of educational contexts. This interpretation remains tentative and warrants further investigation.

Masculinity and Long-term Orientation exhibited pathway-specific moderation across the three relationships, with varying patterns relative to prior studies. This suggests that these cultural dimensions operate through context-dependent mechanisms in different technology adoption domains.

Theoretical and Practical Significance

Theoretically, first, main effects analysis demonstrates that TAM relationships are empirically supported in AIEd, validating TAM as a theoretical framework for understanding in AIEd. Egger’s tests show no evidence of publication bias (Table 3), enhancing confidence in these findings. Second, moderation analysis reveals that contextual factors influence the three core TAM pathways, with some moderators showing consistent directional effects across pathways while others exhibit pathway-specific patterns, indicating that TAM relationships are context-contingent rather than universal. Third, by decomposing broad geographic categorization into specific cultural dimensions and country/region income level factors, this study provides a methodologically refined approach to contextual moderation analysis in TAM research and thereby reveals more granular moderation patterns that geographic classification cannot detect.

Practically, the findings provide evidence-based implications for educators, policymakers, and AI developers. The main effect analyses reveal strong correlations among TAM constructs, with AT-BI demonstrating the strongest relationship, followed by PU-BI and PEU-PU. Implementation efforts should therefore adopt holistic approaches attending to affective, cognitive, and usability dimensions simultaneously. The moderator analyses of the three core pathways (Table 4) reveal systematic variation across contexts. K-12 contexts show consistently stronger relationships than higher education across all three pathways, while students exhibit significantly stronger effects than teachers in the PEU-BI and PU-BI pathways. These patterns suggest that implementations targeting K-12 students can rely more heavily on demonstrating usability and usefulness, whereas approaches for higher education faculty and teachers should recognize that these factors alone may be insufficient and explore complementary strategies tailored to professional contexts. Income-level patterns reveal that ease of use more strongly shapes perceptions of usefulness for lower-income users, whereas usefulness perceptions more strongly drive behavioral intentions for higher-income users. Implementations should therefore prioritize reducing complexity in resource-constrained contexts while emphasizing functional capabilities in well-resourced settings. Cultural dimensions moderate the three core pathways, indicating that cultural factors should be considered when designing and promoting AI in education. In high uncertainty avoidance cultures where TAM relationships are stronger, emphasizing perceived ease of use and usefulness in implementation strategies will be particularly effective. Conversely, weaker effects in high power distance, individualist, and high masculinity cultures indicate that strategies beyond PEU and PU may be necessary.

Limitations and Future Studies

Our database selection prioritized four major indexed databases (Scopus, Web of Science, ERIC, IEEE Xplore) that are widely used in educational technology meta-analyses. While this approach ensures inclusion of high-quality, peer-reviewed studies with sufficient methodological rigor for meta-analytic synthesis, we acknowledge that databases such as PsycINFO and JSTOR were not included, which may have resulted in the exclusion of some relevant studies. Additionally, non-indexed journals and gray literature were not systematically searched, which may limit the generalizability of our findings to the broader landscape of AI acceptance research.

In terms of methodological choices, several considerations warrant acknowledgment. Due to the limited number of studies, Pearson correlation coefficients were primarily used; however, as TAM applications in AIEd expand, future reviews should report both path and correlation coefficients for a more robust analysis. Additionally, the study examined TAM relationships independently and did not consider interrelationships among variables. Consequently, future studies should investigate TAM relationships using techniques such as meta-analytic structural equation modeling. Furthermore, while the categorical moderator approach ensured methodological consistency and interpretability, it may oversimplify continuous constructs such as cultural dimensions. This dichotomization could mask nonlinear relationships that continuous meta-regression might capture. Future meta-analyses with larger study samples could incorporate both categorical and continuous approaches to validate these findings.

Regarding theoretical framework scope, this study excluded variables from related frameworks such as UTAUT (e.g., Performance Expectancy, Effort Expectancy). While these constructs are conceptually similar to PU and PEU, they were not included to avoid potential inconsistency in construct definitions. Future studies could conduct a comparative meta-analysis across TAM and UTAUT to assess their relative explanatory power in AIEd, which would further enrich the understanding of theoretical diversity in this field.

Concerning the scope of moderator and relationship analyses, the study considered a limited set of moderating factors, and future meta-analyses should explore additional variables to provide a more comprehensive understanding of TAM in AIEd. Specifically, this study only conducted moderation effect analyses on the relationships between PEU, PU, and BI, indicating that future research should conduct moderation analyses on a broader range of relationships between TAM constructs. These limitations necessitate future empirical work to validate and extend the current findings across different technological and educational contexts.

Furthermore, most of the studies included in this meta-analysis were conducted during or shortly after the COVID-19 pandemic. During this period, the widespread shift to online and hybrid education significantly accelerated the adoption of AI-based technologies. Consequently, users’ perceptions of usefulness, ease of use, and behavioral intention might have been temporarily elevated due to pandemic-driven necessity rather than long-term voluntary adoption. Therefore, the observed effect sizes in this study may reflect an amplified acceptance pattern influenced by the pandemic context.

In light of the limitations discussed above, Table S4 outlines 35 hypotheses derived from the meta-analytic findings. These propositions warrant further empirical examination to verify whether the observed TAM relationships and moderation patterns persist across specific AIEd implementation contexts.

Conclusions

This meta-analysis provides a comprehensive quantitative synthesis of TAM in the domain of AIEd, addressing a critical gap in TAM research on AI acceptance in educational settings. The findings support TAM’s applicability as a framework for understanding AI acceptance while revealing significant contextual contingencies. Strong positive correlations across TAM constructs (PEU-PU: r = .63; PEU-BI: r = .56; PU-BI: r = .69; AT-BI: r = .72) demonstrate that these core relationships remain significant in AIEd contexts, extending empirical evidence from general educational technology to AI-specific applications in education. Importantly, the substantial heterogeneity observed across all relationships (I² > 95%) and systematic moderation patterns reveal that TAM-based predictions are context-contingent rather than universal. Cultural dimensions, educational levels, user roles, and country/region-level income moderate the strength of core TAM relationships. These findings underscore that understanding AI acceptance through the TAM lens requires careful consideration of contextual factors, crucial for researchers employing this framework and practitioners designing TAM-informed implementation strategies. This study focuses specifically on TAM-based research and does not encompass alternative acceptance frameworks or factors beyond TAM’s core constructs. Within this scope, the findings indicate that while TAM provides a valuable foundation for understanding AI acceptance, its application must be context-sensitive and may require theoretical supplementation in contexts where its explanatory power is weaker. By systematically identifying the contexts in which TAM relationships are stronger or weaker, this study advances more theoretically informed, context-sensitive approaches to understanding and implementing AI technologies in diverse educational settings.

Supplemental Material

sj-docx-1-sgo-10.1177_21582440251409441 – Supplemental material for Technology Acceptance Model in Artificial Intelligence in Education: A Meta-Analysis

Supplemental material, sj-docx-1-sgo-10.1177_21582440251409441 for Technology Acceptance Model in Artificial Intelligence in Education: A Meta-Analysis by Liangyong Xue, Jazihan Mahat and Norliza Ghazali in SAGE Open

Footnotes

ORCID iDs

Liangyong Xue

Jazihan Mahat

Funding

The authors received no financial support for the research,authorship,and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.

Supplemental Material

Supplemental material for this article is available online.

References

Abdullah

Ward

(2016). Developing a general extended technology acceptance model for e-learning (GETAMEL) by analysing commonly used external factors. Computers in Human Behavior, 56, 238–256. https://doi.org/10.1016/j.chb.2015.11.036

Abu-Shanab

E. A.

(2011). Education level as a technology adoption moderator [Conference session]. 2011 3rd International Conference on Computer Research and Development (pp. 324–328). https://doi.org/10.1109/ICCRD.2011.5764029

Ajzen

(1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-T

Al Darayseh

(2023). Acceptance of artificial intelligence in teaching science: Science teachers’ perspective. Computers and Education: Artificial Intelligence, 4, Article 100132. https://doi.org/10.1016/j.caeai.2023.100132

Al Shamsi

J. H.

Al-Emran

Shaalan

(2022). Understanding key drivers affecting students’ use of artificial intelligence-based voice assistants. Education and Information Technologies, 27(6), 8071–8091. https://doi.org/10.1007/s10639-022-10947-3

Albayati

(2024). Investigating undergraduate students’ perceptions and awareness of using ChatGPT as a regular assistance tool: A user acceptance perspective study. Computers and Education: Artificial Intelligence, 6, Article 100203. https://doi.org/10.1016/j.caeai.2024.100203

Al-Emran

Granić

(2021). Is it still valid or outdated? A bibliometric analysis of the technology acceptance model and its applications from 2010 to 2020. In Al-Emran

Shaalan

(Eds.), Recent advances in technology acceptance models and theories (pp. 1–12). Springer International Publishing. https://doi.org/10.1007/978-3-030-64987-6_1

Al-Emran

Mezhuyev

Kamaludin

(2018). Technology acceptance model in M-learning context: A systematic review. Computers & Education, 125, 389–412. https://doi.org/10.1016/j.compedu.2018.06.008

Algerafi

M. A. M.

Zhou

Alfadda

Wijaya

T. T.

(2023). Understanding the factors influencing higher education students’ intention to adopt artificial intelligence-based robots. IEEE Access 11, 99752–99764. https://ieeexplore.ieee.org/document/10247529/

10.

Al-Momani

A. M.

Ramayah

(2024). Adoption of artificial intelligence in education: A systematic literature review. In Al-Sharafi

M. A.

Al-Emran

Tan

G. W.-H.

Ooi

K.-B.

(Eds.), Current and future trends on intelligent technology adoption: Volume 2 (pp. 117–135). Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-61463-7_7

11.

Alomary

Woollard

(2015, November 21). How is technology accepted by users? A review of technology acceptance models and theories [Conference session]. 5th International Conference on 4E. https://eprints.soton.ac.uk/382037/

12.

Alrishan

A. M. H.

(2023). Determinants of intention to use ChatGPT for professional development among Omani EFL pre-service teachers. International Journal of Learning, Teaching and Educational Research, 22(12), 187–209. https://doi.org/10.26803/ijlter.22.12.10

13.

Alshammari

S. H.

Rosli

M. S.

(2020). A review of technology acceptance models and theories. Innovative Teaching and Learning Journal (ITLJ), 4(2), 12–22.

14.

Awal

Md. R.

Haque

Md. E.

(2024). Revisiting university students’ intention to accept AI-powered chatbot with an integration between TAM and SCT: A South Asian perspective. Journal of Applied Research in Higher Education, 17(2), 594–608. https://doi.org/10.1108/JARHE-11-2023-0514

15.

Ayanwale

M. A.

Molefi

R. R.

(2024). Exploring intention of undergraduate students to embrace chatbots: From the vantage point of Lesotho. International Journal of Educational Technology in Higher Education, 21(1), 20. https://doi.org/10.1186/s41239-024-00451-8

16.

Bilquise

Ibrahim

Salhieh

S. M.

(2024). Investigating student acceptance of an academic advising chatbot in higher education institutions. Education and Information Technologies, 29(5), 6357–6382. https://doi.org/10.1007/s10639-023-12076-x

17.

Bond

Khosravi

De Laat

Bergdahl

Negrea

Oxley

Pham

Chong

S. W.

Siemens

(2024). A meta systematic review of artificial intelligence in higher education: A call for increased ethics, collaboration, and rigour. International Journal of Educational Technology in Higher Education, 21(1), 4. https://doi.org/10.1186/s41239-023-00436-z

18.

Borenstein

Hedges

L. V.

Higgins

J. P. T.

Rothstein

H. R.

(2021). Introduction to meta-analysis. John Wiley & Sons.

19.

Cavus

Omonayajo

Mutizwa

M. R.

(2022). Technology acceptance model and learning management systems: Systematic literature review. International Journal of Interactive Mobile Technologies, 17(23), 109–124. https://doi.org/10.3991/ijim.v16i23.36223

20.

Chang

S.-H.

Chou

C.-H.

Yang

J.-M.

(2010). The literature review of technology acceptance model: A study of the bibliometric distributions. PACIS 2010 Proceedings. https://aisel.aisnet.org/pacis2010/158

21.

Choi

Jang

Kim

(2023). Influence of pedagogical beliefs and perceived trust on teachers’ acceptance of educational artificial intelligence tools. International Journal of Human–Computer Interaction, 39(4), 910–922. https://doi.org/10.1080/10447318.2022.2049145

22.

Chuttur

(2009). Overview of the technology acceptance model: Origins, developments and future directions. All Sprouts Content (Vol. 9, Issue 37). https://aisel.aisnet.org/sprouts_all/290

23.

Claro

Castro-Grau

Ochoa

J. M.

Hinostroza

J. E.

Cabello

(2024). Systematic review of quantitative research on digital competences of in-service school teachers. Computers & Education, 215, Article 105030. https://doi.org/10.1016/j.compedu.2024.105030

24.

Cohen

(1988). Statistical power analysis for the behavioral sciences (2nd ed.). Routledge. https://doi.org/10.4324/9780203771587

25.

Dai

C.-P.

Pan

Moon

Liu

(2024). Effects of artificial intelligence-powered virtual agents on learning outcomes in computer-based simulations: A meta-analysis. Educational Psychology Review, 36(1), 31. https://doi.org/10.1007/s10648-024-09855-4

26.

Daniels

M. A.

Greguras

G. J.

(2014). Exploring the nature of power distance: Implications for micro- and macro-level theories, processes, and outcomes. Journal of Management, 40(5), 1202–1229. https://doi.org/10.1177/0149206314527131

27.

Davis

F. D.

(1985). A technology acceptance model for empirically testing new end-user information systems: Theory and results [Doctoral dissertation, Massachusetts Institute of Technology]. https://dspace.mit.edu/handle/1721.1/15192

28.

Davis

F. D.

(1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340. https://doi.org/10.2307/249008

29.

Davis

F. D.

Granić

(2024). The technology acceptance model: 30 years of TAM. Springer International Publishing. https://doi.org/10.1007/978-3-030-45274-2

30.

Deci

E. L.

Ryan

R. M.

(1985). Intrinsic motivation and self-determination in human behavior. Springer US. https://doi.org/10.1007/978-1-4899-2271-7

31.

Dehghani

Mashhadi

(2024). Exploring Iranian English as a foreign language teachers’ acceptance of ChatGPT in English language teaching: Extending the technology acceptance model. Education and Information Technologies, 29(15), 19813–19834. https://doi.org/10.1007/s10639-024-12660-9

32.

Derakhshan

Teo

Saeedy Robat

Janebi Enayat

Jahanbakhsh

A. A.

(2024). Robot-assisted language learning: A meta-analysis. Review of Educational Research, 95(4), 747–774. https://doi.org/10.3102/00346543241247227

33.

Doulani

(2019). An assessment of effective factors in technology acceptance model: A meta-analysis study. Journal of Scientometric Research, 7, 153–166. https://doi.org/10.5530/jscires.7.3.26

34.

Egger

Smith

G. D.

Schneider

Minder

(1997). Bias in meta-analysis detected by a simple, graphical test. BMJ, 315, 629–634. https://doi.org/10.1136/bmj.315.7109.629

35.

Esiyok

Gokcearslan

Kucukergin

K. G.

(2025). Acceptance of educational use of AI chatbots in the context of self-directed learning with technology and ICT self-efficacy of undergraduate students. International Journal of Human–Computer Interaction, 41(1), 641–650. https://doi.org/10.1080/10447318.2024.2303557

36.

Feng

G. C.

Lin

Luo

Zhang

(2021). Determinants of technology acceptance: Two model-based meta-analytic reviews. Journalism & Mass Communication Quarterly, 98(1), 83–104. https://doi.org/10.1177/1077699020952400

37.

Fishbein

Ajzen

(1977). Belief, attitude, intention, and behavior: An introduction to theory and research. Philosophy and Rhetoric, 10(2), 130–132.

38.

Gado

Kempen

Lingelbach

Bipp

(2022). Artificial intelligence in psychology: How can we enable psychology students to accept and use artificial intelligence? Psychology Learning and Teaching, 21(1), 37–56. https://doi.org/10.1177/14757257211037149

39.

Goodhue

D. L.

Thompson

R. L.

(1995). Task-technology fit and individual performance. MIS Quarterly, 19(2), 213–236. https://doi.org/10.2307/249689

40.

Granić

(2022). Educational technology adoption: A systematic review. Education and Information Technologies, 27(7), 9725–9744. https://doi.org/10.1007/s10639-022-10951-7

41.

Granić

Marangunić

(2019). Technology acceptance model in educational context: A systematic literature review. British Journal of Educational Technology, 50(5), 2572–2593. https://doi.org/10.1111/bjet.12864

42.

Gupta

Abbas

A. F.

Srivastava

(2022). Technology acceptance model (TAM): A bibliometric analysis from inception. Journal of Telecommunications and the Digital Economy, 10(3), 77–106. https://doi.org/10.3316/informit.664054795107308

43.

Gursoy

Chi

O. H.

Nunkoo

(2019). Consumers acceptance of artificially intelligent (AI) device use in service delivery. International Journal of Information Management, 49, 157–169. https://doi.org/10.1016/j.ijinfomgt.2019.03.008

44.

Hak

van Rhee

Suurmond

(2016). How to interpret results of meta-analysis [SSRN Scholarly Paper]. Social Science Research Network. https://doi.org/10.2139/ssrn.3241367

45.

Higgins

J. P. T.

Thompson

S. G.

(2002). Quantifying heterogeneity in a meta-analysis. Statistics in Medicine, 21(11), 1539–1558. https://doi.org/10.1002/sim.1186

46.

Hofstede

(1980). Culture’s consequences: International differences in work-related values. Sage Publications.

47.

Hofstede

(2011). Dimensionalizing cultures: The Hofstede model in context. Online Readings in Psychology and Culture, 2(1), 1–26. https://doi.org/10.9707/2307-0919.1014

48.

Hofstede Insights. (2024). Country comparison tool. https://www.hofstede-insights.com/country-comparison-tool

49.

Holmes

Bialik

Fadel

(2019). Artificial intelligence in education: Promises and implications for teaching and learning. Center for Curriculum Redesign.

50.

Huang

Chen

Rau

P.-L. P.

(2022). Exploring acceptance of intelligent tutoring system with pedagogical agent among high school students. Universal Access in the Information Society, 21(2), 381–392. https://doi.org/10.1007/s10209-021-00835-x

51.

Jan

Alshare

K. A.

Lane

P. L.

(2024). Hofstede’s cultural dimensions in technology acceptance models: A meta-analysis. Universal Access in the Information Society, 23(2), 717–741. https://doi.org/10.1007/s10209-022-00930-7

52.

Kelly

Kaye

S.-A.

Oviedo-Trespalacios

(2023). What factors contribute to the acceptance of artificial intelligence? A systematic review. Telematics and Informatics, 77, Article 101925. https://doi.org/10.1016/j.tele.2022.101925

53.

King

W. R.

(2006). A meta-analysis of the technology acceptance model. Information & Management, 43(6), 740–755. https://doi.org/10.1016/j.im.2006.05.003

54.

Labadze

Grigolia

Machaidze

(2023). Role of AI chatbots in education: Systematic literature review. International Journal of Educational Technology in Higher Education, 20(1), 56. https://doi.org/10.1186/s41239-023-00426-1

55.

Lai

C. Y.

Cheung

K. Y.

Chan

C. S.

(2023). Exploring the role of intrinsic motivation in ChatGPT adoption to support active learning: An extension of the technology acceptance model. Computers and Education: Artificial Intelligence, 5, Article 100178. https://doi.org/10.1016/j.caeai.2023.100178

56.

Lee

Kozar

K. A.

Larsen

K. R. T.

(2003). The technology acceptance model: Past, present, and future. Communications of the Association for Information Systems, 12, 752–780. https://doi.org/10.17705/1CAIS.01250

57.

Legris

Ingham

Collerette

(2003). Why do people use information technology? A critical review of the technology acceptance model. Information & Management, 40(3), 191–204. https://doi.org/10.1016/S0378-7206(01)00143-4

58.

Lei

Cui

Zhou

(2018). Relationships between student engagement and academic achievement: A meta-analysis. Social Behavior and Personality: An International Journal, 46(3), 517–528. https://doi.org/10.2224/sbp.7054

59.

Liu

Wang

Evans

Correia

A.-P.

(2024). Critical antecedents of mobile learning acceptance and moderation effects: A meta-analysis on technology acceptance model. Education and Information Technologies, 29(15), 20351–20382. https://doi.org/10.1007/s10639-024-12645-8

60.

Liu

G. L.

Darvin

(2024). Exploring AI-mediated informal digital learning of English (AI-IDLE): A mixed-method investigation of Chinese EFL learners’ AI adoption and experiences. Computer Assisted Language Learning, 38(7), 1632–1660. https://doi.org/10.1080/09588221.2024.2310288

61.

Liu

(2024). Measuring EFL learners’ use of ChatGPT in informal digital learning of English based on the technology acceptance model. Innovation in Language Learning and Teaching, 18(2), 125–138. https://doi.org/10.1080/17501229.2023.2240316

62.

Liu

Y.-L. E.

Huang

Y.-M.

(2025). Exploring the perceptions and continuance intention of AI-based text-to-image technology in supporting design ideation. International Journal of Human–Computer Interaction, 41(1), 694–706. https://doi.org/10.1080/10447318.2024.2311975

63.

C. K.

Hew

K. F.

Jong

M. S.

(2024). The influence of ChatGPT on student engagement: A systematic review and future research agenda. Computers & Education, 219, Article 105100. https://doi.org/10.1016/j.compedu.2024.105100

64.

Liu

(2004). The technology acceptance model: A meta-analysis of empirical findings. Journal of Organizational and End User Computing (JOEUC), 16(1), 59–72. https://doi.org/10.4018/joeuc.2004010104

65.

Lei

(2024). The factors influencing teacher education students’ willingness to adopt artificial intelligence technology for information-based teaching. Asia Pacific Journal of Education, 44(1), 94–111. https://doi.org/10.1080/02188791.2024.2305155

66.

Mafara

Abdullahi

(2024). Adopting artificial intelligence (AI) in education: Challenges & possibilities. Asian Journal of Advanced Research and Reports, 18(2), 106–111. https://doi.org/10.9734/ajarr/2024/v18i2608

67.

Marangunić

Granić

(2015). Technology acceptance model: A literature review from 1986 to 2013. Universal Access in the Information Society, 14(1), 81–95. https://doi.org/10.1007/s10209-014-0348-1

68.

Marikyan

Papagiannidis

Stewart

(2023). Technology acceptance research: Meta-analysis. Journal of Information Science. Advance online publication. https://doi.org/10.1177/01655515231191177

69.

Masa’deh

Majali

S. A.

Alkhaffaf

Thurasamy

Almajali

Altarawneh

Al-Sherideh

Altarawni

(2024). Antecedents of adoption and usage of ChatGPT among Jordanian university students: Empirical study. International Journal of Data and Network Science, 8(2), 1099–1110. https://doi.org/10.5267/j.ijdns.2023.11.024

70.

Mogaji

Viglia

Srivastava

Dwivedi

Y. K.

(2024). Is it the end of the technology acceptance model in the era of generative artificial intelligence? International Journal of Contemporary Hospitality Management, 36(10), 3324–3339. https://doi.org/10.1108/IJCHM-08-2023-1271

71.

Mortenson

M. J.

Vidgen

(2016). A computational literature review of the technology acceptance model. International Journal of Information Management, 36(6, Part B), 1248–1259. https://doi.org/10.1016/j.ijinfomgt.2016.07.007

72.

Mugo

Njagi

Chemwei

Motanya

(2017). The Technology Acceptance Model (TAM) and its application to the utilization of mobile learning technologies. British Journal of Mathematics & Computer Science, 20(4), 1–8. https://doi.org/10.9734/BJMCS/2017/29015

73.

Mustafa

A. S.

Garcia

M. B.

(2021). Theories integrated with technology acceptance model (TAM) in online learning acceptance and continuance intention: A systematic review [Conference session]. 2021 1st Conference on Online Teaching for Mobile Education (OT4ME) (pp. 68–72). https://doi.org/10.1109/OT4ME53559.2021.9638934

74.

Cheung

(2023). Understanding secondary students’ continuance intention to adopt AI-powered intelligent tutoring system for English learning. Education and Information Technologies, 28(3), 3191–3216. https://doi.org/10.1007/s10639-022-11305-z

75.

Oliver

R. L.

Linda

(1981). Effect of satisfaction and its antecedents on consumer preference and intention. ACR North American Advances, NA-08. https://www.acrwebsite.org/volumes/9746/volumes/v08/NA-08/full

76.

Ouyang

Zhang

Xue

Rashid

A. M.

Pyng

H. S.

Hassan

A. B.

(2025). The cultural compass: A systematic review on cultural dimensions theory in educational settings. SAGE Open, 15(2), 21582440251342160. https://doi.org/10.1177/21582440251342160

77.

Page

M. J.

McKenzie

J. E.

Bossuyt

P. M.

Boutron

Hoffmann

T. C.

Mulrow

C. D.

Shamseer

Tetzlaff

J. M.

Moher

(2021). Updating guidance for reporting systematic reviews: Development of the PRISMA 2020 statement. Journal of Clinical Epidemiology, 134, 103–112. https://doi.org/10.1016/j.jclinepi.2021.02.003

78.

Peterson

R. A.

Brown

S. P.

(2005). On the use of beta coefficients in meta-analysis. Journal of Applied Psychology, 90(1), 175–181. https://doi.org/10.1037/0021-9010.90.1.175

79.

Pillai

Sivathanu

Metri

Kaushik

(2024). Students’ adoption of AI-based teacher-bots (T-bots) for learning in higher education. Information Technology and People, 37(1), 328–355. https://doi.org/10.1108/ITP-02-2021-0152

80.

R Core Team. (2023). R: A language and environment for statistical computing [Computer software]. https://www.R-project.org/

81.

Rahman

M. S.

Sabbir

M. M.

Zhang

Moral

I. H.

Hossain

G. M. S.

(2023). Examining students’ intention to use ChatGPT: Does trust matter? Australasian Journal of Educational Technology, 39(6), 51–71. https://doi.org/10.14742/ajet.8956

82.

Rogers

E. M.

(1962). Diffusion of innovations. Free Press of Glencoe. NY, 32, 891–937.

83.

Rosli

M. S.

Saleh

N. S.

Md. Ali

Abu Bakar

Mohd Tahir

(2022). A systematic review of the technology acceptance model for the sustainability of higher education during the COVID-19 pandemic and identified research gaps. Sustainability, 14(18), Article 11389. https://doi.org/10.3390/su141811389

84.

Sabah

N. M.

(2016). Exploring students’ awareness and perceptions: Influencing factors and individual differences driving m-learning adoption. Computers in Human Behavior, 65, 522–533. https://doi.org/10.1016/j.chb.2016.09.009

85.

Schepers

Wetzels

(2007). A meta-analysis of the technology acceptance model: Investigating subjective norm and moderation effects. Information & Management, 44(1), 90–103. https://doi.org/10.1016/j.im.2006.10.007

86.

Scherer

Siddiq

Tondeur

(2019). The technology acceptance model (TAM): A meta-analytic structural equation modeling approach to explaining teachers’ adoption of digital technology in education. Computers & Education, 128, 13–35. https://doi.org/10.1016/j.compedu.2018.09.009

87.

Silva

(2007). Post-positivist review of technology acceptance model. Journal of the Association for Information Systems, 8(4), 255–266. https://doi.org/10.17705/1jais.00121

88.

Strzelecki

ElArabawy

(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. https://doi.org/10.1111/bjet.13425

89.

Sukkeewan

Songkram

Nasongkhla

(2024). Investigating students’ behavioral intentions towards a smart learning platform based on machine learning: A user acceptance and experience perspective. International Journal of Information and Education Technology, 14(2), 260–270. https://doi.org/10.18178/ijiet.2024.14.2.2047

90.

Sun

Tian

Sun

Fan

Yang

(2024). Pre-service teachers’ inclination to integrate AI into STEM education: Analysis of influencing factors. British Journal of Educational Technology, 55(6), 2574–2596. https://doi.org/10.1111/bjet.13469

91.

Suurmond

Van Rhee

Hak

(2017). Introduction, comparison, and validation of Meta-Essentials: A free and simple tool for meta-analysis. Research Synthesis Methods, 8(4), 537–553. https://doi.org/10.1002/jrsm.1260

92.

Tao

Wang

Zhang

(2020). A systematic review and meta-analysis of user acceptance of consumer-oriented health information technologies. Computers in Human Behavior, 104, Article 106147. https://doi.org/10.1016/j.chb.2019.09.023

93.

Tarhini

Hone

Liu

(2014). The effects of individual differences on e-learning users’ behaviour in developing countries: A structural equation model. Computers in Human Behavior, 41, 153–163. https://doi.org/10.1016/j.chb.2014.09.020

94.

Tiwari

C. K.

Bhat

Mohd. A.

Khan

S. T.

Subramaniam

Khan

M. A. I.

(2023). What drives students toward ChatGPT? An investigation of the factors influencing adoption and usage of ChatGPT. Interactive Technology and Smart Education, 21(3), 333–355. https://doi.org/10.1108/ITSE-04-2023-0061

95.

Turner

Kitchenham

Brereton

Charters

Budgen

(2010). Does the technology acceptance model predict actual use? A systematic literature review. Information and Software Technology, 52(5), 463–479. https://doi.org/10.1016/j.infsof.2009.11.005

96.

UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO.

97.

Vaccaro

Almaatouq

Malone

(2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8(12), 2293–2303. https://doi.org/10.1038/s41562-024-02024-1

98.

Venkatesh

Bala

(2008). Technology acceptance model 3 and a research agenda on interventions. Decision Sciences, 39(2), 273–315. https://doi.org/10.1111/j.1540-5915.2008.00192.x

99.

Venkatesh

Davis

F. D.

(2000). A theoretical extension of the technology acceptance model: Four longitudinal field studies. Management Science, 46(2), 186–204. https://doi.org/10.1287/mnsc.46.2.186.11926

100.

Venkatesh

Morris

M. G.

Davis

G. B.

Davis

F. D.

(2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425–478. https://doi.org/10.2307/30036540

101.

Venkatesh

Thong

J. Y. L.

(2012). Consumer acceptance and use of information technology: Extending the unified theory of acceptance and use of technology. MIS Quarterly, 36(1), 157. https://doi.org/10.2307/41410412

102.

Viechtbauer

(2010). Conducting meta-analyses in R with the metafor package. Journal of Statistical Software, 36, 1–48. https://doi.org/10.18637/jss.v036.i03

103.

Viechtbauer

Cheung

M. W.-L.

(2010). Outlier and influence diagnostics for meta-analysis. Research Synthesis Methods, 1(2), 112–125. https://doi.org/10.1002/jrsm.11

104.

World Bank. (2022). New World Bank country classifications by income level: 2022-2023. World Bank Blogs. https://blogs.worldbank.org/en/opendata/new-world-bank-country-classifications-income-level-2022-2023

105.

Zhao

Zhu

Tan

Zheng

(2011). A meta-analysis of the impact of trust on technology acceptance model: Investigation of moderating influence of subject and context type. International Journal of Information Management, 31(6), 572–581. https://doi.org/10.1016/j.ijinfomgt.2011.03.004

106.

(2024). Do AI chatbots improve students learning outcomes? Evidence from a meta-analysis. British Journal of Educational Technology, 55(1), 10–33. https://doi.org/10.1111/bjet.13334

107.

(2024). Unraveling effects of AI chatbots on EFL learners’ language skill development: A meta-analysis. Asia-Pacific Education Researcher, Advance online publication. https://doi.org/10.1007/s40299-024-00853-2

108.

Xue

Ghazali

Mahat

(2025). Artificial intelligence (AI) adoption among teachers: A systematic review and agenda for future research. International Journal of Technology in Education, 8(3), 802–824. https://doi.org/10.46328/ijte.1191

109.

Xue

Mahat

Ghazali

Shi

(2025). Technology acceptance model (TAM) in artificial intelligence in higher education (AIHEd): A systematic literature review. Journal of Institutional Research South East Asia, 23(1), 355–375.

110.

Xue

Rashid

A. M.

Ouyang

(2024). The unified theory of acceptance and use of technology (UTAUT) in higher education: A systematic review. Sage Open, 14(1), 21582440241229570. https://doi.org/10.1177/21582440241229570

111.

Yousafzai

S. Y.

Foxall

G. R.

Pallister

J. G.

(2007a). Technology acceptance: A meta-analysis of the TAM: Part 1. Journal of Modelling in Management, 2(3), 251–280. https://doi.org/10.1108/17465660710834453

112.

Yousafzai

S. Y.

Foxall

G. R.

Pallister

J. G.

(2007b). Technology acceptance: A meta-analysis of the TAM: Part 2. Journal of Modelling in Management, 2(3), 281–304. https://doi.org/10.1108/17465660710834462

113.

(2023). A meta-analysis of the effect of virtual reality technology use in education. Interactive Learning Environments, 31(8), 4956–4976. https://doi.org/10.1080/10494820.2021.1989466

114.

Yucel

U. A.

Gulbahar

(2013). Technology acceptance model: A review of the prior predictors. Ankara University Journal of Faculty of Educational Sciences (JFES), 46(1), 89–109. https://doi.org/10.1501/Egifak_0000001275

115.

Zawacki-Richter

Marín

V. I.

Bond

Gouverneur

(2019). Systematic review of research on artificial intelligence applications in higher education—Where are the educators? International Journal of Educational Technology in Higher Education, 16(1), 39. https://doi.org/10.1186/s41239-019-0171-0

116.

Zhang

Schießl

Plößl

Hofmann

Gläser-Zikuda

(2023). Acceptance of artificial intelligence among pre-service teachers: A multigroup analysis. International Journal of Educational Technology in Higher Education, 20(1), 49. https://doi.org/10.1186/s41239-023-00420-7

117.

Zhang

Weng

Zhu

(2018). The relationships between electronic banking adoption and its antecedents: A meta-analytic study of the role of national culture. International Journal of Information Management, 40, 76–87. https://doi.org/10.1016/j.ijinfomgt.2018.01.015

118.

Zhao

Pinto Llorente

A. M.

Sánchez Gómez

M. C.

(2021). Digital competence in higher education research: A systematic literature review. Computers & Education, 168, Article 104212. https://doi.org/10.1016/j.compedu.2021.104212

119.

Zheng

Niu

Zhong

Gyasi

J. F.

(2023). The effectiveness of artificial intelligence on learning achievement and learning perception: A meta-analysis. Interactive Learning Environments, 31(9), 5650–5664. https://doi.org/10.1080/10494820.2021.2015693

120.

Zou

Lyu

Han

Zhang

(2025). Exploring students’ acceptance of an artificial intelligence speech evaluation program for EFL speaking practice: An application of the integrated model of technology acceptance. Computer Assisted Language Learning, 38(5–6), 1366–1391. https://doi.org/10.1080/09588221.2023.2278608

121.

Zou

Huang

(2023). To use or not to use? Understanding doctoral students’ acceptance of ChatGPT in writing through technology acceptance model. Frontiers in Psychology, 14, Article 1259531. https://doi.org/10.3389/fpsyg.2023.1259531

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.09 MB