Extending the Technology Acceptance Model for Mathematics Education: AI Literacy,AI-TPACK,and Self-Efficacy Among Pre-Service Teachers

Abstract

This research aims to investigate pre-service mathematics teachers’ (PsMT) intention to use artificial intelligence (AI) technologies. Specifically, this study modified the technology acceptance model (TAM) for this investigation by including variables such as AI literacy, AI-TPACK, and AI self-efficacy. Utilizing a correlation cross-sectional design was the objective of the research. In Türkiye, data were collected from 696 PsMT at 13 universities and analyzed with covariance-based structural equation modeling. The results showed that intention to use AI is primarily determined by perceived usefulness, perceived ease of use, and attitude. Furthermore, AI literacy was less likely to be a predictor of behavioral intention than an antecedent that supports proximal constructs such as perceived usefulness, AI-TPACK, and self-efficacy. AI-TPACK provided support for perceived ease of use and self-efficacy, but not for perceived usefulness or behavioral intention directly. As a consequence, self-efficacy has helped students see AI tools as less intimidating and has cultivated a more positive attitude towards them. The findings indicate that acceptance of AI is a multidimensional process encompassing cognitive, pedagogical, and psychological dimensions, underscoring the importance of comprehensive learning experiences in Teacher Education Institutions that address these dimensions.

Keywords

AI literacy self-efficacy AI-TPACK technology acceptance model pre-service mathematics teacher artificial intelligence in education mathematics education

Introduction

Artificial intelligence (AI)’s impact on social and economic life has increased significantly in recent years, owing to improved access and a growing range of tools (European Commission, 2023). The ability to interact successfully with AI has become a fundamental skill, requiring individuals to learn new techniques and reconsider their roles (Carolus et al., 2023; Wang et al., 2023).

The increasing influence of AI can be seen especially in education, where teaching methods have rapidly changed through the use of AI tools (Zhang & Aslan, 2021). Advances in deep learning and neural network architecture-based AI can be used to personalise learning, provide real-time feedback, predict performance, and assist the teacher’s actions (Chassignol et al., 2018; Chen et al., 2020; Yoon et al., 2024).

Mathematics education benefits from the considerable pedagogical potential of AI-supported tools. Mathematics is a subject that, by its very nature, requires abstraction, logic, and step-by-step instructions. Students often struggle to relate concrete representations to abstract mathematical ideas and to address conceptual gaps. As a result, students often view mathematics as a difficult subject (Chin et al., 2022; Gravemeijer, 2005; Mangarin & Caballes, 2024). Nonetheless, standard teaching methods may not always ensure immediate and adaptive support to remove such barriers. AI tools can support lessons by tailoring to student needs, providing individual, real-time feedback, and presenting information interactively (Selwyn, 2022; Son, 2024; Wu et al., 2023). More specifically, a transformation of static objects in geometry, algebra, and calculus into more interactive forms, such as AR/VR, dynamic mathematics software, or interactive visualization, can improve students’ attitudes toward mathematics and enhance their conceptual engagement (Awang et al., 2025; Gadanidis, 2017; Mohamed et al., 2022).

Even though the technology is available, a lot will depend on teachers, especially pre-service teachers (PSTs), who will shape future classrooms. The implementation of technology for educational purposes depends on its acceptance and adoption (Ma & Lei, 2024; Nikolopoulou et al., 2021; Pal & Patra, 2021). Since teacher-unsupported innovations rarely survive (Lin & Van Brummelen, 2021), it is important to understand the factors affecting PSTs’ intentions to adopt AI tools.

The Technology Acceptance Model (TAM) (Davis, 1989) is one of the most used models to explain an individual’s intentions to accept technology via core constructs perceived usefulness (PU) and perceived ease of use (PEOU) and attitude (ATT) (e.g. Khong et al., 2023; Teo, 2011; Zhang et al., 2023) The implementation of AI in schools depends not only on various cognitive acceptance variables, i.e. whether the teacher finds the tool useful and easy to use, but also on PSTs’ self-efficacy regarding AI, their competence to integrate those technologies pedagogically, and their AI literacy (e.g. Al Abdullatif, 2024; An et al., 2023; Ma & Lei, 2024).

Therefore, the current study broadens the TAM to incorporate self-efficacy (SE), AI-TPACK and AI Literacy. In this framework, SE refers to PSTs’ belief in their ability to use AI tools in instruction (Chou et al., 2024). In contrast, AI-TPACK indicates their capacity to employ and integrate AI tools in mathematics instruction in pedagogically meaningful ways (Celik, 2023). AI literacy is a primary form of readiness. It describes the ability to understand and evaluate AI systems and use them in an ethically informed way (Long & Magerko, 2020; Ng et al., 2021).

Although some of these constructs had been studied individually or in limited combinations, there has been little research that addresses these variables in a single model and considers their interrelationships in relation to AI, especially for pre-service mathematics teachers (e.g., Al Abdullatif, 2024; Choi et al., 2023; Li, 2025). This gap is particularly significant in the context of mathematics education, as the teaching of mathematics involves a structure that places high cognitive demands due to abstract concepts and complex problem-solving processes (Phan et al., 2017; Wakhata et al., 2023). In this context, Celik (2023) and Santos-Trigo (2024), for instance, held that their functional characteristics do not exclusively determine the educational value of these tools. However, they also have pedagogically appropriate uses to support conceptual understanding and reasoning, problem-solving, and real-time feedback. Use of these tools requires PSTs to have not only strong self-efficacy beliefs about the effectiveness of these tools, but also sufficient AI literacy; that is, the ability to understand, critically evaluate, and use AI in an ethical way (Long & Magerko, 2020; Ng et al., 2021).

In light of this theoretical and empirical gap, the current study proposes an integrated model of pre-service mathematics teachers’ (PSMTs) intention to adopt AI tools in Mathematics, showing how AI literacy, SE, and AI-TPACK together contribute to this intention. In essence, the study seeks to offer a more context-specific and theory-informed account of AI adoption in teacher education.

Theoretical Framework and Hypothesis Development

TAM Constructs

The Technology Acceptance Model (TAM) based on Fishbein and Ajzen’s Theory of Reasoned Action (TRA, 1975) explains the individuals’ intention to adopt (Davis, 1989). Perceived Usefulness (PU), Perceived Ease of Use (PEOU), Attitude Toward Use (ATT) and Behavioural Intention (BI) are the four constructs under the study. PU indicates that using a system will improve one’s performance while PEOU relates to the sense that it requires little effort (Davis, 1989). The way we perceive something impacts how we feel about it. Studies have shown that what they believe about education affects teachers’ integration of digital technologies into instruction (Al Abdullatif, 2024; Eickelmann & Vennemann, 2017; Gurer and Akkaya, 2021; Ibili et al., 2019; Mailizar et al., 2021).

The proposed model of TAM has expanded beyond its original boundaries through the introduction of TAM2 (Venkatesh & Davis, 2000), UTAUT (Venkatesh et al., 2003), and TAM3 (Venkatesh & Bala, 2008). Nonetheless, PU, PEOU, and ATT remain important predictors of technology use (Marangunić & Granić, 2015). Empirical research has shown that PU and PEOU are powerful determinants of teachers’ attitudes or intentions towards technology use (Al Darayseh, 2023; Gado et al., 2022; Zhang et al., 2023) and that PU has the more powerful of the two effects (Baydaş & Göktaş, 2017; Teo et al., 2009; Teo, 2012). In the TAM, PEOU enhances attitude and intention by lowering perceived effort. In contrast, PU enhances attitude and intention by increasing expected performance (Davis, 1989). It has been consistently observed that these pathways exist in educational contexts (Al Darayseh, 2023; Gado et al., 2022). In this study, TAM has used not as primary source of theoretical originality, but as a fundamental conceptual framework enabling the assessment of the explanatory contribution of AI-specific variables. Consequently, the inclusion of classical TAM relaationships in the model aims not so much to revalidate TAM, but rather to highlight the additional contribution made by constructs such as AI literacy, AI-TPACK and self-efficacy in explaining PSTs’ intention to adopt AI tools.

Based on these theoretical and empirical foundations, the following hypotheses are proposed:

H1a

PsMTs’ PU of AI technologies significantly predicts their BI to use them.

H1b

PsMTs’ PU of AI technologies significantly predicts their ATT toward using them.

H2a

PsMTs’ PEOU of AI technologies significantly predicts their BI to use them.

H2b

PsMTs’ PEOU of AI technologies significantly predicts their ATT toward using them.

H2c

PsMTs’ PEOU of AI technologies significantly predicts their PU.

PsMTs’ ATT toward AI technologies significantly predict their BI to use them

The Role of AI Literacy

In the 21st century, AI literacy is an essential digital competence that prepares people for future societal needs (Lee et al., 2021; Long & Magerko, 2020; Yang, 2022). Kandlhofer et al. (2016) coined the term to highlight the importance of understanding the principles underlying AI technologies. Currently, AI literacy refers to a set of competencies that enable individuals to use AI tools competently and responsibly; to think critically and creatively about AI systems; and to participate in digital society actively (Kong et al., 2022; Ng et al., 2022; Wang et al., 2023). While AI technologies can form part of a broader umbrella of digital technologies, AI literacy differs from digital literacy. This is because AI draws on a wide range of disciplines, including computer science, mathematics, psychology, sociology, linguistics, and philosophy (Russell & Norvig, 2003). In fact, much like computer and digital literacy, AI literacy does not require people to gain expertise in AI theories or technology; instead, it focuses on the ability to use AI products competently and responsibly (Wang et al., 2023).

Educational contexts have become the focal point of AI-literacy discussions. Integrating AI into teaching has heightened interest in this competency (Wang et al., 2023). Within this framework, in the context of teacher education, PSTs are expected not only to adopt emerging technologies but also to identify appropriate AI tools, use them ethically, and evaluate their suitability for instructional purposes.

At this point, it is important to clarify the conceptual distinction between AI literacy and AI-TPACK. While AI-TPACK represents a specialized professional framework that enables teachers to determine how to structure the affordances of AI (e.g., personalized learning pathways, automated assessment, or student monitoring) for pedagogical purposes within their discipline (Ning et al., 2024; Celik, 2023), AI literacy involves informing individuals, at a general level, about the functioning, use, evaluation, and ethical dimensions of AI systems. Therefore, in this study, AI literacy refers to PSMTs’ ability to recognize, use, evaluate, and engage with AI technologies critically and ethically.

Research shows that AI literacy improves learners’ knowledge, attitudes, and confidence in using AI (Du et al., 2024; Su & Yang, 2024). This kind of competence is particularly important in mathematics education, where PSMTs are expected to identify an AI tool that meets their learning goals, use it ethically, and evaluate an AI-generated explanation, representation, and solution in terms of their mathematical accuracy, reliability, and pedagogical appropriateness. According to this framework, AI literacy is assumed to directly determine PSMTs’ self-efficacy, AI-TPACK, PU, PEOU, and BI. Thus, AI literacy is deemed a precursor construct that shapes their cognitive, pedagogical, and behavioral orientation toward AI.

In accordance with this conception, AI literacy also influences PU as it enables users to understand AI’s advantages and limitations, evaluate tools critically and appreciate their advantages (Laupichler et al., 2022) If teachers develop a better understanding of AI concepts, they can acknowledge its potential in learning and teaching – such as personalized learning and efficient assessment – thereby increasing its perceived usefulness (Jang, 2024; Pan et al., 2025; Qi et al., 2025). Conversely, other studies did not establish a significant association (Al Abdullatif, 2024).

AI literacy may also shape PEOU. Teachers with higher AI literacy generally find AI systems more intuitive, experience fewer difficulties, and perceive them as less complex (Al Abdullatif, 2024; Pan et al., 2025). With a higher level of literacy, one can understand what the system does. It will help the customer use and adopt the system effectively (Jang, 2024; Qi et al., 2025). Contextual moderation may be occurring, according to some studies that report no direct effect (Al Abdullatif, 2024).

Beyond PEOU, AI literacy may also help in the development of an AI-TPACK, which enables teachers to link AI tools with pedagogy and content knowledge (Celik, 2023). The foundational understanding provided by AI Literacy guides its practical application. Improving the technological aspect of TPACK, AI literacy helps teachers select tools in alignment with pedagogical goals and maximise their instructional value (Kong et al., 2022). According to Al Abdullatif (2024), teachers who are proficient in AI literacy have better understanding which enables them to use AI tools more meaningfully in bigger contexts.

Also, the self-efficacy of PSTs in artificial intelligence use can be strengthened through AI literacy. According to Bećirović et al. (2025) and Lim (2023), AI literacy is a strong predictor of self-efficacy because digital literacy enhances teachers’ self-efficacy. AI literacy combines subject knowledge, practical skills and pedagogical knowledge to enable PSTs to apply AI tools effectively. Integrating AI into instruction involves developing teachers’ capacity and has become increasingly essential.

In conclusion, these cognitive, pedagogical, and self-efficacy–based effects of AI literacy may also be reflected in PSMTs’ intentions to adopt and use artificial intelligence. In this regard, AI literacy determines individuals’ readiness to adopt AI (Wang et al., 2023) and predicts teachers’ effective use of AI in classrooms (Jang, 2024; Schiavo et al., 2024). Still, findings are mixed: some report no direct link to behavioral intention (Wang et al., 2025), whereas others reveal indirect effects (Qi et al., 2025).

H4a

PsMTs’AI literacy levels significantly predict their BI to use AI technologies

H4b

PsMTs’AI literacy levels significantly predict their PEOU of AI technologies.

H4c

PsMTs’AI literacy levels significantly predict their PU of AI technologies.

H4d

PsMTs’AI literacy levels significantly predict their AI-TPACK levels.

H4e

PsMTs’AI literacy levels significantly predict their AI self-efficacy.

The Role of AI-TPACK

TPACK, which refers to teachers’ competence in integrating technology into instruction (Koh et al., 2013), has evolved considerably since Shulman introduced the concept of pedagogical content knowledge. By merging technological, pedagogical, and content knowledge, it supports the design of effective teaching strategies (Mishra et al., 2023; Mishra & Koehler, 2006). In mathematics education, TPACK plays a crucial role in integrating digital technology due to its flexibility across various contexts and tools (e.g., Kadıoğlu-Akbulut et al., 2023; Li & Nugraha, 2025; Sofyan et al., 2023).

The rise of AI in education has prompted a reevaluation of the intersection of technology, pedagogy, and content (Ning et al., 2024). To capture teachers’ competence in this new context, the framework has been extended as AI-TPACK (Ning et al., 2024; Celik, 2023). AI-TPACK assesses teachers’ ability to select, apply, and integrate AI tools in line with pedagogical aims (Hava & Babayiğit, 2025; Celik, 2023). At this point, it is important to distinguish AI-TPACK from AI literacy. While AI literacy refers to a more general readiness to understand, evaluate, and use artificial intelligence ethically, AI-TPACK specifically denotes the capacity to integrate AI tools into the instructional process in pedagogically meaningful ways.

Recent research identifies TPACK and the TAM as complementary perspectives for examining the integration of digital technology in education (Hsu, 2016; Joo et al., 2018; Li, 2025; Mailizar et al., 2021). While TPACK focuses on the interaction of technological, pedagogical, and content knowledge (Mishra et al., 2023), TAM explains teachers’ perceptions and attitudes toward adopting technologies, such as online teaching platform or AI-based tools (Al Abdullatif, 2024; Khong et al., 2023). Within this framework, an educator who is competent in AI-TPACK may be more inclined to perceive AI tools as more accessible and easier to use. In addition, as they develop a clearer understanding of how these tools can be integrated with instructional strategies, they are better able to evaluate their concrete contributions to supporting instructional processes, enhancing student engagement, and improving performance, and thus perceive them as more useful (Al Abdullatif, 2024; Runge et al., 2025).

Empirical studies show that teachers’ or PSTs’ AI-TPACK levels significantly predict PEOU, PU, and overall acceptance (An et al., 2023; Sun et al., 2024). However, results across contexts are inconsistent. Runge et al. (2025) revealed that PSTs’ AI-TPACK levels had a direct effect on PEOU and PU of AI technologies but no significant effect on behavioral intention. Al Abdullatif (2024) reported that AI-TPACK had a direct effect on PEOU, but did not have a significant effect on PU and BI. In contrast, Tram (2025) and Runge et al. (2025) reported that AI-TPACK significantly influenced both PU and PEOU.

These contextual variations suggest the need for further research to explore how AI-TPACK affects PU, PEOU, and BI. This issue becomes even more critical in the context of mathematics education. Effective use of AI tools requires competence in how these tools can be pedagogically integrated with abstract mathematical concepts, multiple representations, and problem-solving processes (Getenet, 2024; Wardat et al., 2023). Such competence is expected to strengthen pre-service mathematics teachers’ understanding of how and for what purposes AI tools can be integrated into instructional processes. In this regard, AI-TPACK levels are considered likely to be associated with pre-service teachers’ intentions to use AI tools.

Self-efficacy is also linked to AI-TPACK. Drawing on Bandura’s (1986) social cognitive theory, which posits that competence and mastery experiences in a given domain are among the key determinants of self-efficacy beliefs, it can be argued that AI-TPACK competence may enhance teachers’ confidence and perceived self-efficacy in using AI tools (Tram, 2025). Teachers with stronger AI-TPACK typically show higher confidence in using AI, and this confidence directly enhances self-efficacy (Sun et al., 2024; Tram, 2025). However, Xu et al. (2025) found that self-efficacy affected AI-TPACK, suggesting that the direction of the relationship may vary across contexts.

Accordingly, the following hypotheses were proposed:

H5a

PsMTs’AI-TPACK levels significantly predict their PU of AI technologies.

H5b

PsMTs’AI-TPACK levels significantly predict their PEOU of AI technologies.

H5c

PsMTs’AI-TPACK levels significantly predict their BI to use AI technologies

H5d

PsMTs’AI-TPACK levels significantly predict their AI self-efficacy.

The Role of Self-Efficacy

Self-efficacy (SE) refers to an individual’s belief in their ability to plan and execute actions to achieve specific goals (Bandura, 1977, 1986). Context-dependent in nature (Bandura, 1986), it has been examined across various domains, including computer, internet, and technology self-efficacy (e.g. Hong et al., 2019; Nardi & Ranieri, 2019; Tondeur et al., 2019; Yildiz Durak, 2018). This construct is a fundamental concept for understanding motivational processes across various contexts, including education. Teacher self-efficacy, as an adaptation of this belief to the instructional context, reflects teachers’ beliefs in their ability to achieve desired outcomes in the teaching process (Bandura, 1977; Tschannen-Moran & Hoy, 2001).

Consistent with previous studies, the concept of “self-efficacy for AI use in teaching” is also grounded in Bandura’s (1977) self-efficacy framework. This concept refers to teachers’ beliefs and self-appraisal capacities regarding their ability to effectively use AI in instructional processes in order to achieve improved educational outcomes (Chou et al., 2024; Guo et al., 2025; Yang, 2025). In this regard, AI self-efficacy is conceptualized in the present study as a belief construct that reflects PSMTs’ perceived competence in using AI tools in teaching and integrating these tools into classroom practices, and that may influence their perceptions of and attitudes toward AI.

Mathematics education, by its nature, requires the concretization of abstract concepts, and AI tools such as intelligent tutoring systems and chatbots can provide personalized learning experiences in this regard (Dermeval et al., 2018; Shin, 2022). However, teachers’ ability to integrate complex algorithms, systems, and tools into classroom settings depends primarily on their self-efficacy in using these technologies (Li, 2024).

Pre-service teachers with higher levels of AI self-efficacy are more likely to perceive the use of these technologies as less complex and more manageable (Zhang et al., 2023). However, evidence remains inconsistent: while some studies report strong direct effects of self-efficacy on ease of use (Al Adwan et al., 2023; Al Darayseh, 2023; Kong et al., 2024), others find self-efficacy affects behavioral intention rather than ease of use (Tram, 2025), or even that PEOU predicts self-efficacy (Sun et al., 2024).

Within the framework of Self-Determination Theory, individuals’ sense of competence plays a fundamental role in the development of intrinsic motivation and positive attitudes toward an activity (e.g., Chiu, 2022). When pre-service teachers feel confident in their ability to use AI tools effectively for instructional purposes, they may develop more positive attitudes toward this technology. (Kong et al., 2024). Self-efficacy also directly influences attitude (Kong et al., 2024), although findings in this regard are likewise inconsistent (Falebita and Kok., 2025). Further research is therefore needed to clarify these links in AI-based contexts.

Therefore, the following hypotheses are proposed:

H6a

PsMTs’AI self-efficacy significantly predicts their PEOU of AI technologies.

H6b

PsMTs’AI self-efficacy significantly predicts their attitudes ATT toward AI technologies.

Figure 1 illustrates the proposed research model, grounded in the theoretical foundations and hypotheses outlined above.

Figure 1.

Proposed model

Methods

Research Design

This study employed a quantitative, cross-sectional, correlational survey design. This study investigated PSMTs’ intentions to use AI technologies in mathematics instruction and how AI-TPACK, AI literacy, and AI self-efficacy predict these intentions. It also aimed to model the interrelations among these factors within an extended TAM framework. All ethical principles were adhered to within the scope of the study, and informed consent was obtained from the participants. The research was conducted with the approval of the Istanbul Medeniyet University Social and Human Sciences Ethics Committee.

Participants

The sample for this study consisted of 696 pre-service mathematics teachers enrolled at 13 universities in Türkiye, selected through convenience sampling. The dataset initially consisted of 732 individuals. Prior to the main analyses, several data screening procedures were conducted to ensure the quality of the dataset. No missing data were observed because questionnaires with missing responses were excluded during data collection. Moreover 30 cases were removed due to careless or patterned responding.

Univariate outliers were examined by calculating z-scores for the observed scores; values with |z| > 3.29 were indentified. Mahalanobis distance (D2) values were used to test for multivariate outliers; the critical value for df = 7 at p < .001 is 24.32. Given these criteria, several responses were identified as potential multivariate outliers. In light of the above cases, the dataset for six subjects was excluded. Consequently, responses from 696 persons formed the final dataset. RMSEA approach (MacCallum et al., 1996); model df = 360; power analysis indicates power approximately equal to 1.00. Showing that the sample size was sufficient to detect model misspecification. This analysis is one of the major post-hoc power confirmations. Participants consisted of 74.9% females (n = 521) and 25.1% males (n = 175), which reflects the typical gender distribution within the teaching profession. Regarding grade level, 14.1% (n = 98) were first-year students, 22.6% (n = 157) were second-year students, 28% (n = 195) were third-year students, and 35.3% (n = 246) were fourth-year students. While 70.4% (n = 490) reported some experience with AI, only 15.2% (n = 106) had received formal training, suggesting limited structured exposure despite widespread informal familiarity with technology.

Instruments

The survey, grounded in the extended TAM, consisted of two sections. The first part of the questionnaire includes demographic information such as sex, grade level, and whether or not one has experienced AI before or attended an AI training. The items in the subsequent section were employed to assess PEOU, PU, ATT, BI, SE, AI-TPACK, and AIL constructs in the context of AI.

In developing the measurement instrument, first define the theoretical boundaries of each construct, then survey the scales that have appeared in the relevant literature, with supporting validity and reliability evidence. In this sense, we selected items related to the AIL and AI-TPACK constructs from existing scales that have been previously adapted into Turkish. Items for the PU, PEOU, ATT, BI, and SE constructs were developed by conceptually and contextually adapting statements from validated scales reported in the literature to align with the context of artificial intelligence and instruction. During the process, whilst some items have adapted directly to be context, others have rewritten to suit the context of AI and education, based on the conceptual content of the relevant items. For items without a Turkish version, the original English statements were first adapted to the research context and then translated into Turkish. The translation process used the translation–back-translation technique. The original English items were translated into Turkish by an academic expert in English language studies. Then, the items were back-translated into English by an independent expert proficient in English and Turkish. The researcher and two bilingual experts compared the back-translated versions with the original items to assess semantic equivalence, linguistic consistency, and conceptual appropriateness. Appropriate changes were executed as needed.

The measurement tools prepared in this way are then proposed by three mathematics education experts, measurement and evaluation experts, and educational technology experts through evaluations of content validity and the clarity of reasoning and language in the research context. At the wording level, three items were revised at the expert’s recommendation. The revised version was then trialed on a sample of 25 pre-service mathematics teachers with characteristics similar to those of the main sample. After the pilot study, the items’ clarity, applicability, and response process were reassessed to arrive at the final version of the instrument. The confirmatory factor analysis was applied to the final measurement model. In addition, validity and reliability were assessed using relevant statistical indicators. For detailed information on source scales, sample items, see Appendix 1.

Artificial Intelligence Literacy (AIL): The six AI literacy items used in the study were derived from the AI literacy scale originally developed by Wang et al. (2023) and adapted into Turkish by Çelebi et al. (2023).

Specifically, six of the eleven items in the Turkish adaptation were used in the present study. An example item is: “I can identify the AI technologies employed in the applications and products I use.” The reliability coefficient was α = .897, the composite reliability (CR) was .897, and the average variance extracted (AVE) was .593, all of which meet the accepted psychometric standards for reliability and construct validity.

Artificial Intelligence Technological Pedagogical Content Knowledge (AI-TPACK): This construct consists of five items derived from the AI-TPACK scale originally developed by Ning et al. (2024) and adapted into Turkish by Canbazoğlu Bilici et al. (2024). The selected items were also contextually adapted to the mathematics content. An example item is: “I am capable of selecting appropriate teaching methods, and AI applications based on mathematics content for instruction.” The reliability coefficient (α) was 0.832, the CR was 0.840, and the AVE was 0.516, all of which indicate satisfactory levels of reliability and convergent validity for the measurement model.

Perceived Usefulness (PU): Items related to the PU construct were conceptually and contextually adapted based on Davis’s (1989) PU framework and the TAM applications reported by Şimşek et al. (2025). The items were reworded to align with the use of AI tools in educational contexts, while maintaining the core conceptual content of the constructs they were intended to measure. For example, the statement “Using [system] improves the quality of the work I do” in Davis’s (1989) scale was adapted in this study as “AI tools can improve the quality of my teaching. The reliability coefficient was α = .753, the CR was .750, and the AVE was .500, indicating that the construct was at an acceptable level in terms of reliability and validity.

Perceived Ease of Use (PEOU): Items related to the PEOU construct were conceptually and contextually adapted based on Davis’s (1989) framework for PEOU, taking into account subsequent applications of this framework, particularly those by Teo (2011) and Venkatesh and Bala (2008). The items were modified to reflect the use of AI tools in an educational setting, while retaining the essential conceptual content of the constructs they were intended to measure. For example, the item “I find the system to be easy to use” in Venkatesh and Bala’s (2008) scale was adapted in this study as “Using AI tools is easy for me.” The reliability coefficient was α = .841, the composite reliability (CR) was .844, and the average variance extracted (AVE) was .575, all of which meet the established thresholds for measurement reliability and validity.

Attitude Toward AI (ATT): Items related to the ATT construct were conceptually and contextually adapted from relevant statements in previous studies addressing attitudes toward technology use. In this context, the studies of Shirvani (2014), Caratiquit and Caratiquit (2022), and Aksekili and Kan (2024) were considered. The items were modified to reflect the use of AI tools in an educational setting, while retaining the essential conceptual content of the constructs they were intended to measure. For example, the statement “I like using computers for teaching purposes” was adapted in this study as “I like using AI tools in mathematics teaching”. The reliability coefficient (Cronbach’s alpha) was α = .857, the composite reliability (CR) was .864, and the average variance extracted (AVE) was .614, all of which indicate strong reliability and convergent validity.

Behavioral Intention (BI): Items related to the BI construct were conceptually and contextually adapted from relevant statements in prior studies addressing the intention to use of technology and AI. In this regard, the studies by Ayanwale et al., 2022; Miranda & Chamorro-Mera, 2025; Venkatesh & Bala, 2008 were considered. The items were reworded to reflect the intention to use AI technology in future mathematics teaching. For example, the statement “I plan to use the system in the next <n> months” was adapted in this study as “I plan to use AI technology in my future mathematics lessons.” The reliability coefficient was α = .875, the composite reliability (CR) was .876, and the average variance extracted (AVE) was .638, all of which exceed the commonly accepted thresholds for reliability and construct validity.

Self-Efficacy (SE): Items related to the SE construct were conceptually and contextually adapted from relevant statements in prior studies examining individuals’ self-efficacy beliefs regarding AI and technology. In this process, the studies by Wang et al. (2025), Schwarzer &Jerusalem. (1995), and Kwon et al. (2019) were considered. The items were modified to reflect the use of AI tools in an educational setting, while retaining the essential conceptual content of the constructs they were intended to measure. For example, the statement “I am confident I can use AI technology well” was contextually adapted in this study as “I am confident I can use AI technologies in my teaching”. The reliability coefficient was α = .793, the composite reliability (CR) was .792, and the average variance extracted (AVE) was .560, all of which exceed the commonly accepted thresholds for reliability and construct validity.

Data Collection and Analysis

Data collection took place over approximately four weeks in 2025. The participation was voluntary and online informed consent was obtained from all participants. The survey was conducted using google forms. It would take about 10–15 mins to fill the form.

To verify the proposed research model CB-SEM was adopted because the core objective of the study was to test a theoretically established model based on TAM. CB-SEM is suitable for Confirming a Theory since it allows the researcher to assess the global model fit indices after which the researcher also gets to estimate the relationships among the latent constructs in a Confirmatory manner (Hair et al., 2019; Kline, 2023).

Model parameters were analyzed using the statistical software R (version 4.5.1), and SEM computations were conducted with the lavaan package. IBM SPSS Statistics 20 was also used. Following Anderson & Gerbing's (1988) two-step approach, the measurement model was first assessed for reliability and validity, followed by testing the structural model and estimating direct, indirect, and total effects. This approach, widely used in prior studies (Gurer & Akkaya, 2022; Al Adwan et al., 2023; Kong et al., 2024), enables the refinement of measurement properties prior to hypothesis testing.

The measurement model was evaluated using confirmatory factor analysis (CFA). To evaluate the model–data fit, the following indices were used: χ², χ²/df, p-value, CFI, TLI, RMSEA, RMSEA %90 confidence interval, and SRMR. In addition, the scaling correction factor obtained in the model estimation was 1.46. Since MLR was used for model estimation, scaled χ² and the robust versions of CFI, TLI, and RMSEA, along with the RMSEA 90% confidence interval, were reported. CFI and TLI values ≥ 0.95 indicate an excellent fit, and values ≥ 0.90 indicate an acceptable fit (Hu & Bentler, 1999). When RMSEA and SRMR values are below 0.06 and 0.08, respectively, they indicate an excellent model fit, while values below 0.08 and 0.10 are considered acceptable (Schreiber et al., 2006). To assess the reliability and convergent validity of the measurement model, Cronbach’s alpha, composite reliability (CR), and average variance extracted (AVE) values were calculated. Discriminant validity was examined using the heterotrait–monotrait ratio (HTMT) and Fornell-Lacker Criterion. In addition, measurements were taken for multicollinearity and common method bias.

In the structural model analysis, bootstrapping with 10,000 resamples was performed in order to obtain more stable confidence intervals (Falk & Biesanz, 2015). The criterion for the significance of a mediation effect was that the 95% confidence interval did not include the value zero (0). (MacKinnon et al., 2004; Preacher & Selig, 2012).

Results

This section presents the findings obtained from the analyses conducted within the scope of the research. The results of the research are presented according to the sub-problems and the theoretical framework established in this study. The initial stage presents an overview of the model, including its graphical presentation and basic statistics, then the model assessment, and finally the outcome.

Preliminary Data Analysis

Descriptive Statistics

Table 1 shows the means, standard deviations, skewness, and kurtosis of the study variables. All constructs had mean scores above the midpoint of the 5-point Likert scale (3.0), indicating overall positive responses. The minor standard deviations (0.56–0.79) indicate limited variability across participants, while the skewness (−0.80 to 0.04) and kurtosis (−0.07 to 0.65) values fall within acceptable ranges (Kline, 2015).

Table 1.

Descriptive Statistics and Normality Estimates

Variable	Mean	SD	Skewness	Kurtosis
BI	3.94	0.60	−0.27	0.24
PU	3.81	0.60	−0.17	0.42
PEOU	4.04	0.56	−0.16	0.19
SE	3.44	0.68	−0.07	−0.07
ATT	3.89	0.60	−0.19	0.20
AI-TPACK	4.05	0.79	−0.80	0.16
AIL	3.20	0.61	0.04	0.65

In addition to univariate normality, multivariate normality was assessed using the Mardia coefficient. The results showed that the values of multivariate skewness (16716.68, p < .001) and kurtosis (134.11, p < .001) were statistically significant, indicating that the data deviate from a multivariate normal distribution. To address this violation, the robust maximum likelihood estimation method (MLR) was used to ensure parameter estimates were less sensitive to the normality assumption.

Multicollinearity and Common Method Bias

Some assumptions related to the predictor variables were examined before the analysis. The potential multicollinearity problem among the predictor variables was evaluated using the Variance Inflation Factor (VIF) values. The results indicated that the VIF values ranged from 1.56 to 2.21. Since these values are well below the recommended threshold of 3 in the literature (Hair et al., 2019), it was concluded that there is no multicollinearity problem in the model.

In addition to statistical assumptions, common-method bias was examined. For this purpose, Harman’s single-factor test was conducted. The analysis results indicated that the first factor explains 34.46% of the total variance. The absence of a single dominant factor and the finding that a single factor does not account for the majority of the variance suggest that common method bias is not a serious concern (Podsakoff et al., 2003).

Evaluation of the Measurement Model

Convergent Validity

Convergent validity was evaluated through factor loadings, composite reliability (CR), average variance extracted (AVE), and Cronbach’s alpha (Table 2). All standardized loadings were statistically significant (p < .001) and generally exceeded the .70 threshold, ranging from .584 to .841. Factor loading of .50 or higher is generally considered practically significant, as they indicate that at least 25% of the variance in the observed variable is explained by the underlying factor (Hair et al., 2019). Internal consistency was confirmed, as both α and CR values for all constructs were above .70. Moreover, AVE values for most constructs exceeded the .50 benchmark, while the AVE value for PU was at the threshold (.500); however, since its CR (.752) was well above .70, convergent validity was still considered adequate (Fornell & Larcker, 1981).

Table 2.

Internal and Convergent Assessment

Construct	Item	Loading (λ)	Cronbach’s alpha (α)	CR	AVE
AI-TPACK	AI-TPACK1	0.584	0.832	0.840	0.516
	AI-TPACK2	0.723
	AI-TPACK3	0.677
	AI-TPACK4	0.804
	AI-TPACK5	0.784
AI Literacy	AIL1	0.808	0.897	0.897	0.593
	AIL2	0.745
	AIL3	0.740
	AIL4	0.777
	AIL5	0.803
	AIL6	0.748
PEOU	PEOU1	0.781	0.841	0.844	0.575
	PEOU2	0.813
	PEOU3	0.726
	PEOU4	0.709
PU	PU1	0.718	0.753	0.750	0.500
	PU2	0.679
	PU3	0.724
SE	SE1	0.776	0.793	0.792	0.560
	SE2	0.718
	SE3	0.751
ATT	AT1	0.700	0.857	0.864	0.614
	AT2	0.841
	AT3	0.826
	AT4	0.759
BI	BI1	0.827	0.875	0.876	0.638
	BI2	0.815
	BI3	0.765
	BI4	0.788

Discriminant Validity

In Table 3, the discriminant validity of the constructs was evaluated using the heterotrait–monotrait ratio (HTMT) (Franke & Sarstedt, 2019; Henseler et al., 2015), which is considered a more stringent and modern criterion compared to traditional methods. All HTMT ratios among the constructs were below the conservative threshold of .85, with only one exception (PU-BI: .853). However, the literature widely recognizes that for conceptually related constructs, HTMT values between .85 and .90 are still considered acceptable (Henseler et al., 2015). Therefore, all constructs in the model successfully met the criteria for discriminant validity.

Table 3.

Discriminant Validity-HeteroTrait-MonoTrait Ratio of Correlations

	AI-TPACK	PEOU	AIL	ATT	SE	PU
PEOU	0.432
AIL	0.664	0.277
ATT	0.491	0.729	0.402
SE	0.620	0.338	0.608	0.452
PU	0.493	0.684	0.417	0.813	0.458
BI	0.479	0.762	0.371	0.834	0.372	0.853

Table 4 shows that the square root of the AVE values for all constructions is higher than their correlations with other constructions. This indicates that discriminant validity at an acceptable level within the framework of the Fornell–Larcker criterion. However, the relatively high correlations between certain construct pairs, such as BI–ATT and BI–PU, are theoretically expected because PU is positioned as a key cognitive evaluation of ATT and BI within the TAM framework (e.g., Davis, 1989; Falebita & Kok, 2025), and because ATT is one of the closest predictors of BI (e.g., Al Darayseh & Mersin, 2025; Davis, 1989).

Table 4.

Discriminant Validity Evaluation: Fornell–Lacker Criterion

	BI	PU	PEOU	SE	ATT	AIL	AI-TPACK
BI	.799
PU	.691	.707
PEOU	.654	.543	.758
SE	.318	.364	.290	.748
ATT	.723	.651	.619	.391	.784
AIL	.334	.346	.242	.516	.362	.770
AI-TPACK	.417	.394	.373	.506	.421	.578	.718

Note. Bold figures are the square root of AVEs.

The convergent and discriminant validity results together confirmed that the measurement model had adequate construct validity. The obtained values for the measurement model were χ² (356) = 588.64, (p < .001), CFI = .969, TLI = .964, RMSEA = .037 (90% CI [.032, .042]), and SRMR = .043. According to these values, the model–data fit is acceptable (Hu & Bentler, 1999). Overall, the model showed satisfactory fit and met the validity requirements for subsequent structural analysis.

Evaluation of the Structural Model

Following the assessment of the measurement model, the structural model was tested to evaluate the hypothesized relationships among the constructs. The structural model demonstrated an excellent fit to the data: Χ² (360) = 596.721 (p < .001), Χ²/df = 1.65, CFI = .968, TLI = .964, RMSEA = .037 [%90 CI: .032, .042], and SRMR = .043. The consistency of these indices with the measurement model values indicates that the structural constraints did not significantly degrade the model fit. The model’s explanatory power was substantial. The R² values indicate that the model explains 82.0% of the variance in BI and 70.8% of the variance in ATT. Additionally, the model accounted for 54.8% of the variance in PU, 46.3% in SE, 41.8% in AI-TPACK, and 21.3% in PEOU. According to Chin’s (1998) criteria, the model demonstrated substantial explanatory power for BI and ATT, moderate explanatory power for PU, SE, and AI-TPACK, and weak explanatory power for PEOU. These findings suggest that the model was particularly strong in explaining pre-service teachers’ BI and Att toward AI use, but relatively limited in explaining perceived ease of use.

Table 5 presents the standardized regression coefficients (β), standard errors (SE), z-values, and p-values of the direct effects. According to the analysis results, 13 of the 17 proposed hypotheses were supported, while 4 (H4a, H4b, H5a, and H5c) were rejected.

Table 5.

Structural Paths Assessment

Hypothesis	Path	β	SE	z	p	Results
H1a:	PU→BI	0.505	0.114	5.082	<.001	Accepted
H1b:	PU→ATT	0.562	0.080	7.340	<.001	Accepted
H2a:	PEOU→BI	0.196	0.072	3.134	0.002	Accepted
H2b:	PEOU→ATT	0.291	0.068	4.505	<.001	Accepted
H2c:	PEOU→PU	0.597	0.062	9.737	<.001	Accepted
H3:	ATT→BI	0.267	0.100	2.938	0.003	Accepted
H4a:	AIL→BI	−0.021	0.042	−0.503	0.615	Rejected
H4b:	AIL→PEOU	−0.090	0.062	−1.271	0.204	Rejected
H4c:	AIL→PU	0.196	0.050	3.515	<.001	Accepted
H4d:	AIL→AI-TPACK	0.646	0.062	11.537	<.001	Accepted
H4e:	AIL→SE	0.377	0.057	6.963	<.001	Accepted
H5a:	AI-TPACK→PU	0.100	0.051	1.562	0.118	Rejected
H5b:	AI-TPACK→PEOU	0.400	0.068	4.718	<.001	Accepted
H5c:	AI-TPACK→BI	0.041	0.041	0.927	0.354	Rejected
H5d:	AI-TPACK→SE	0.374	0.059	5.987	<.001	Accepted
H6a:	SE→PEOU	0.167	0.060	2.331	0.02	Accepted
H6b:	SE→ATT	0.104	0.045	2.068	0.039	Accepted

AIL had a significant effect on AI-TPACK (β = .646, z = 11.537, p < .001). In addition, AIL positively and significantly influenced SE (β = .377, z = 6.963, p < .001) and PU (β = .196, z = 3.515, p < .001). Moreover, AI-TPACK had significant effects on SE (β = .374, z = 5.987, p < .001) and PEOU (β = .400, z = 4.718, p < .001).

Within the dimensions of the TAM, PEOU significantly predicted PU (β = .597, z = 9.737, p < .001), ATT (β = .291, z = 4.505, p < .001), and BI (β = .196, z = 3.134, p = 0.002). SE also had positive and significant effects on both PEOU (β = .167, z = 2.331, p = 0.020) and ATT (β = .104, z = 2.068, p = 0.039). Regarding BI, PU (β = .505, z = 5.082, p < .001) and ATT (β = .267, z = 2.938, p = 0.003) had significant effects.

However, four direct effects were not statistically significant. Specifically, AIL did not significantly influence BI (H7: β = −.021, p = 0.615) or PEOU (H8: β = −.090, p = 0.204). Similarly, the direct effects of AI-TPACK on PU (H5a: β = .100, p = 0.118) and BI (H5c: β = .041, p = 0.354) were not statistically significant.

As shown in Table 6, AIL demonstrated significant direct effects on AI-TPACK (β = .646, p < .001), SE (β = .377, p < .001), and PU (β = .196, p = .001). However, its direct effects on PEOU (β = −.090, p = .208) and BI (β = −.021, p = .618) were not significant. Despite the non-significant direct effects, AIL exerted significant indirect effects on PEOU (β = .322, p < .001), ATT (β = .347, p < .001), PU (β = .166, p = .003), and BI (β = .270, p < .001). Consequently, the total effects of AIL were .232 on PEOU, .362 on PU, and .249 on BI, all statistically significant (p < .001). Similarly, AI-TPACK significantly influenced SE (β = .374, p < .001) and PEOU (β = .400, p < .001) directly. Although its direct effects on PU (β = .100, p = .124) and BI (β = .041, p = .365) were not significant, AI-TPACK showed significant indirect effects on PU (β = .239, p < .001), ATT (β = .346, p < .001), and BI (β = .291, p < .001). As a result, the total effects of AI-TPACK were .339 on PU and .332 on BI (p < .001).

Table 6.

Direct, Indirect, and Total Effects in the Model

Dependent variable	Independent variable	Direct effect	95% CI	Indirect effect	95% CI	Total effect(β)
BI	PU	0.505	[.368, .842]	0.150	[.052, .307]	0.655
ATT	PU	0.562	[.445, .772]	‐	‐	0.562
BI	PEOU	0.196	[.080, .367]	0.469	[.419, .694]	0.666
ATT	PEOU	0.291	[.164, .440]	0.336	[.250, .487]	0.627
PU	PEOU	0.597	[.483, .728]	‐	‐	0.597
BI	ATT	0.267	[.083, .500]	‐	‐	0.267
BI	AIL	−0.021	[-.105, .062]	0.270	[.189, .382]	0.249
PEOU	AIL	−0.090	[-.206, .043]	0.322	[.193, .384]	0.232
PU	AIL	0.196	[.077, .273]	0.166	[.049, .245]	0.362
AI-TPACK	AIL	0.646	[.595, .842]	‐	‐	0.646
SE	AIL	0.377	[.282, .508]	0.242	[.172, .345]	0.619
ATT	AIL	‐	‐	0.347	[.240, .418]	0.347
PU	AI-TPACK	0.100	[-.024, .184]	0.239	[.114, .283]	0.339
PEOU	AI-TPACK	0.400	[.192, .458]	‐	‐	0.400
BI	AI-TPACK	0.041	[-.046, .120]	0.291	[.170, .393]	0.332
SE	AI-TPACK	0.374	[.243, .480]	‐	‐	0.374
ATT	AI-TPACK	‐	‐	0.346	[.196, .402]	0.346
PEOU	SE	0.167	[.021, .259]	‐	‐	0.167
ATT	SE	0.104	[.007, .186]	0.105	[.013, .179]	0.209
BI	SE	‐	‐	0.111	[.032, .187]	0.111

Within the TAM framework, PEOU had significant direct effects on PU (β = .597, p < .001), ATT (β = .291, p < .001), and BI (β = .196, p = .002). In addition, PEOU exerted a strong indirect effect on BI (β = .469, p < .001), yielding a substantial total effect of .666 (p < .001). SE significantly influenced PEOU (β = .167, p = .021) and ATT (β = .104, p = .042) directly. Moreover, SE had a significant indirect effect on ATT (β = .105, p = .028) and BI (β = .111, p = .006), leading to total effects of .209 on ATT and .111 on BI. Finally, PU emerged as a key predictor in the model, showing significant direct effects on ATT (β = .562, p < .001) and BI (β = .505, p < .001). In addition, PU had a significant indirect effect on BI through attitude (β = .150, p = .007), resulting in a total effect of .655 (p < .001). Consistent with the TAM framework, ATT also had a significant direct effect on BI (β = .267, p = .005).

Figure 2 illustrates the path diagram and the regression coefficients associated with the tested model.

Figure 2.

Sem diagram

Discussion

According to the findings of this study, the use of AI in mathematics instruction cannot be justified solely based on the classical components of TAM, i.e., PU and PEOU. Additionally, the influence of PSMTs’ AI literacy, their knowledge of incorporating AI into the instructional process (AI-TPACK), and self–efficacy perceptions on the AI use. In this regard, the study offers a more elaborate explanatory framework that pulls together the cognitive, pedagogical, and psychological dimensions in the context of mathematics teacher education, thereby contributing to the TAM literature. Plus, the model provides a strong explanation of Behavioral Intention (R² = .82). This value is notably higher than the R² values reported by Li and Manzari (2025) (.633) and Li (2025) (.642). A plausible reason for this stronger explanatory performance is that the current model incorporates AI-related constructs, including AI literacy, AI-TPACK, and self-efficacy, which appear to enhance the prediction of behavioral intention.

Core TAM Relationships in AI Acceptance

The findings regarding to H1a–H3 indicate that PU, PEOU and ATT are complementary core mechanisms in PSMTs adoption of AI technology. This means their intention to use AI technologies is influenced by factors beyond technical ease. They also see technology as suitable for teaching mathematics, improving students’ knowledge, and facilitating problem-solving. They also have positive feelings towards these technologies. The original TAM and subsequent extensions (Al Darayseh, 2023; Davis, 1989; Falebita & Kok, 2025) remain supported by this pattern and the underlying assumptions.

To begin with, the significant influence of PU on both BI and attitude means PSMTs regard AI tools as useful only if they think these tools could make a concrete contribution to maths teaching. The finding is consistent with earlier studies emphasizing the importance of PU in accepting technologies (Sun et al., 2024; Zhao et al., 2025; Şimşek et al., 2025). The contributions could be interpreted as functions that clarify abstract mathematical concepts, diagnose students’ misconceptions, assist students in solving problems, and provide feedback.

In addition, the significant effects of PEOU on BI, ATT, and PU suggest that PSMTs who perceive AI tools as easier to use are more likely to evaluate them as useful for mathematics instruction, develop more positive orientations toward them, and demonstrate a greater tendency to adopt them. This finding is also consistent with the literature on the acceptance of educational technologies. Prior studies have shown PEOU influences PU (Al Adwan et al., 2023; Al Darayseh, 2023), ATT (Al Darayseh, 2023; Şimşek et al., 2025), BI (Tram, 2025; Zhao et al., 2025). In domains such as mathematics instruction, which require high levels of cognitive and pedagogical planning, tools perceived as complex are less likely to be evaluated as both useful and positively received.

Finally, the significant predictive effect of positive attitudes toward AI on BI indicates that PSMTs’ favorable affective and evaluative orientations toward AI tools strengthen their intention to use them. In other words, perceiving these tools as valuable and engaging increases the likelihood of adoption (Falebita & Kok, 2025; Kong et al., 2024; Li, 2025).

AI Literacy as a Foundational Antecedent of AI Acceptance

The findings regarding AI literacy indicate that this construct functions not as a proximal variable that directly explains AI acceptance, but rather as a foundational antecedent that supports the acceptance process across cognitive, pedagogical, and psychological perspectives. While the direct effects of AI literacy on BI and PEOU were not significant, its effects on PU, AI-TPACK, and self-efficacy were significant. The significant and strong effect of AI literacy on AI-TPACK suggests that as PSMTs’ knowledge and awareness of AI increase, their competence in integrating these technologies pedagogically into mathematics instruction also improves, which is consistent with similar findings reported in studies conducted in other contexts (Al Abdullatif, 2024; Aldemir et al., 2025; Biton & Segal, 2025). In other words, PSMTs who better understand how AI operates, what affordances it offers, and what limitations it entails can evaluate these tools more critically in relation to instructional goals, content, and pedagogical requirements, and to integrate them more appropriately into teaching processes (Celik, 2023; Daher, 2025).

The research finding that AI literacy significantly predicts SE in accordance with Bandura’s (1997) framework of self-efficacy suggests that the more PSMTs’ understand and are aware of AI, the stronger their belief in their capability to use these technologies to teach mathematics will be. This finding is consistent with earlier studies (Becirovic et al., 2025; Du et al., 2024; Ji et al., 2025; Lim, 2023). Additionally, the substantial influence of AI literacy on PU indicates that PSMTs’ knowledge and awareness of AI may enable them to better assess the potential usefulness of these technologies in the context of mathematics instruction, which is consistent findings reported in other contexts (Qi et al., 2025; Schiavo et al., 2024). This result suggests that AI is not only a technical tool but also a resource that supplies tangible inputs to math teaching and is therefore considered useful (e.g., Ma & Lei, 2024; Pan et al., 2025).

AI literacy has no significant impact on PEOU, suggesting that knowledge and awareness of AI do not translate into perceptions of ease of use (Al Abdullatif, 2024). Not only cognitive familiarity, but also procedural experience, practical skills, and perceptions of ease from using the technology can help explain PEOU. The significant indirect effect indicates that while AI literacy does not directly impact users’ ease of use perceptions, it does influence them through more proximal constructs, notably self-efficacy and pedagogical integration competence. As new teachers become more familiar with AI, they will build confidence with these technologies and learn how to integrate them into instruction. This preparedness may lead to AI tools being characterized as more controllable and user-friendly.

As revealed in the study’s findings, AI literacy does not directly affect the intention to use AI. This suggests that knowledge and awareness of AI, alone not sufficient to foster an intention to use it. The assumption that technology can be useful, combined with the belief that one can use it, will form a stronger behavioral intention (Ma & Lei, 2024; Qi et al., 2025). Consequently, it seems more plausible that AI literacy has an indirect effect on BI through more immediate factors PU and self-efficacy.

AI-TPACK as a Pedagogical Pathway to AI Acceptance

The findings related to AI-TPACK suggest that this construct functions not as an outcome variable that directly determines AI acceptance, but rather as a structure that strengthens PSTs’ preparedness for how to use AI pedagogically. Specifically, the significant effects of AI-TPACK on PEOU (H5b) and SE (H5d) indicate that as PSMTs’ knowledge of how to integrate AI into instructional processes increases, they perceive these technologies as more manageable and develop stronger self-efficacy beliefs regarding their effective use.

AI tools can support teachers at several stages of the instructional process, especially in mathematics instruction. For example, during lesson planning, one can produce suitable examples, questions, and multiple representations. During instruction, the teacher can diagnose student errors, provide alternative explanations, and offer solution steps within the problem-solving process. Post instruction, by analyzing student responses and producing personalized feedback. Improved knowledge of pedagogical integration will help PSTs more easily comprehend AI tools’ role in the teaching process, perceive them as more manageable and easier to operate, and enhance their self-efficacy in using them effectively. Recent research supports these effects: Al Abdullatif (2024), Runge et al. (2025), An et al. (2023), and Sun et al. (2024) found significant AI-TPACK impacts on PEOU; Tram (2025) likewise showed that TPACK enhances self-efficacy and fosters innovative pedagogy.

In contrast, the non-significant direct effects of AI-TPACK on PU (H5a) and BI (H5c) suggest that pedagogical integration knowledge alone may not be sufficient to generate a strong perception of the instructional value of these technologies or a direct intention to use them. This may be because AI-TPACK primarily represents a knowledge structure that explains how AI can be used within the instructional process. In contrast, PU and BI are shaped not only by such knowledge but also by beliefs about the technology’s contribution to student learning and by self-efficacy regarding its effective use.

In addition, the relatively recent exposure of PSMTs to AI tools may provide a further explanation for this finding. Although they may be able to use these tools at a technical level, they may not yet possess sufficient experience or practical knowledge regarding when, how, and for which pedagogical purposes these tools should be integrated into authentic instructional contexts. Therefore, having a certain level of knowledge about how to integrate AI into teaching may not always be sufficient to perceive these technologies as useful or to develop a clear intention to use them. This finding is also consistent with results reported in the literature indicating that AI-TPACK has no significant effect on BI (Al Abdullatif, 2024; Li, 2025; Runge et al., 2025) or PU (Al Abdullatif, 2024).

The Role of AI Self-Efficacy in AI Acceptance

The findings indicate that SE plays a significant role in shaping PSMTs’ perceptions of usability and their affective orientations toward AI. The significant effects of SE on PEOU and ATT suggest that self-efficacy beliefs may be a key determinant of perceiving AI tools as more manageable and easier to use, and of developing more positive attitudes toward these technologies. Similar results regarding PEOU were reported by Al Adwan et al. (2023), Falebita and Kok (2025), and Zhang et al. (2023).

Particularly in the context of mathematics instruction, PSMTs’ belief that they can effectively use an AI tool for tasks such as identifying students’ error patterns, generating alternative solution strategies, verifying the mathematical correctness of outcomes, and pedagogically refining AI-generated outputs may support their perception of these technologies as more manageable and easier to use. Similarly, this sense of competence may facilitate evaluating AI tools not as systems that generate uncertainty but as functional resources that support instruction, thereby contributing to the development of more positive attitudes toward these technologies. Similar results regarding ATT were reported by Kong et al. (2024) and Xu et al. (2025).

According to Bandura (1997), SE refers to individuals’ beliefs in their capability to successfully perform a specific task, and these beliefs can influence motivation, effort, persistence, and responses to stress or anxiety. In this regard, SE can be considered a key construct that functions between cognitive evaluations and affective resilience in teachers’ adoption of AI. The indirect effect of self-efficacy on behavioral intention also suggests that this construct influences AI acceptance not directly, but through shaping how PSMTs perceive and evaluate the technology.

Theoretical Contribution

The current research helps the literature on AI acceptance in theoretical terms by incorporating not only the classic TAM components PU, PEOU, and ATT, but also cognitive, pedagogical, and psychological dimensions such as AI literacy, AI-TPACK, and SE. The results show that AI literacy does not serve as a distal variable that directly affects BI. Instead, AI literacy serves as a bedrock antecedent that supports more proximate antecedents of AI acceptance, P. AI, AI-TPACK, and SE.

Similarly, AI-TPACK has a significant effect on PEOU and SE. However, it does not directly affect PU and BI. Hence, the fact that it does not directly mediate acceptance. Rather than functioning as a direct predictor of acceptance, this construct seems to reflect a pedagogically preparatory form of knowledge that informs how AI can be situated and used within teaching contexts. Furthermore, the notably significant relationships between SE, PEOU, and ATT indicate that this construct serves as a crucial psychological bridge between knowledge-based structures and behavioral orientations.

In this respect, the study presents a more refined, multi-dimensional, and theoretically enriched framework of AI acceptance in the context of mathematics teacher education. However, further research can examine the direct and indirect effects of this model’s constructs across other teacher groups and contexts. Longitudinal and experimental designs may reveal more about the relationships among changes in AI literacy, AI-TPACK, and self-efficacy, not only with behavioural intention but also with actual classroom use and pedagogical practices.

Practical Implications

Findings of the study are significant for teacher education programmes and policymakers to enhance applications of AI in schools. First, the findings that PU and PEOU significantly influence BI suggest that merely introducing the technical features of AI tools may not be sufficient for their adoption by PSMTs. Thus, pre-service programs must ensure practice-based experiences that render visible the tangible contributions of AI in mathematics instruction. For instance, students should have example lesson functions that use AI to analyze students’ error patterns, produce alternative solution strategies, provide multiple representations, give personalized feedback, and check the correctness of solutions. In this way, PSMTs may begin to perceive these technologies not merely as novel tools but as pedagogically useful and functional resources.

In addition, the finding that AI literacy supports BI primarily through PU, SE, and pedagogical integration knowledge suggests that AI-related training for PSMTs should aim not only to raise awareness but also to enhance perceptions of instructional value, confidence in use, and pedagogical competence. Accordingly, teacher education programs should equip PSMTs with skills such as effective prompt design, evaluating AI-generated mathematical solutions for accuracy and pedagogical appropriateness, restructuring outputs to consider students’ misconceptions, transitioning across different forms of representation, and adapting content to students’ levels.

Similarly, the significant effects of AI-TPACK on PEOU and SE indicate that PSMTs need practice-oriented learning opportunities in which they can experience using AI tools within instructional processes. Therefore, teacher education programs should include activities such as designing AI-supported lesson plans, microteaching practices, case-based discussions, and the pedagogical evaluation of AI-generated outputs. The absence of a direct effect of AI-TPACK on PU and BI further suggests that pedagogical integration knowledge alone may not lead to adoption; rather, PSMTs need concrete opportunities to experience the actual instructional value of AI in classroom contexts.

Moreover, the significant effects of SE on both PEOU and ATT highlight the critical importance of developing PSMTs’ confidence in their ability to use AI tools. For this reason, teacher education programs should create low-stakes practice spaces for PSMTs to play with AI across different teaching situations, try out different ways to use the tech, work with peers on instructional decisions, and structure reflections on the play. These supported practices are likely to ease the anxiety associated with AI, help to view these technologies as more controllable, and enhance attitudes. At the institutional level, it would also be useful for faculties to prepare sample instructional materials, ethical-use guidelines, and repositories of AI applications for the teaching of mathematics to faculty and pre-service teachers.

Limitations and Future Research

This study provides crucial insights into AI acceptance among PSMTs, but it has its limitations. One, the cross-sectional design of the study precludes us from making causal interpretations about the relationships among the constructs included in the model. While the direct and indirect effects of AI literacy, AI-TPACK, and SE on BI are significant, the evolution of these relationships over time cannot be studied in this research. The second risk involves the use of self-report measures to collect data. This can lead to a number of biases, such as social desirability bias, self-evaluation bias, and an increased risk of over-reporting more favourable intentions toward AI. Further, the research was conducted solely with PSMTs in Türkiye using convenience sampling; thus, the findings cannot be generalized to other subject areas, the in-service teacher population, or even a different cultural and institutional context.

Furthermore, this study focused on individuals’ cognitive, pedagogical, and psychological acceptance of the AI technology. However, it did not take into account contextual factors such as institutional support, technological infrastructure, AI content in the syllabus, opportunities for practice, and instructor guidance. Notably, the finding that AI-TPACK has no direct effect on PU and BI, and that AI literacy has no direct effect on BI, indicates that these constructs may function differently across contexts. Consequently, future studies should examine the model across different groups of teachers and educational contexts and apply longitudinal, experimental designs. Mixed-methods approaches to explore the extent to which changes in AI literacy, AI-TPACK, and self-efficacy are related not only to BI but also to actual classroom use and pedagogy. The incorporation of self-reports can be supplemented with other data collection methods, such as observations, performance tasks, digital trace data, and qualitative interviews, to provide a better understanding of what AI acceptance looks like in practice.

Conclusions

The use of AI by PSMTs will not be influenced solely by the classical TAM constructs, such as PU, PEOU, and ATT. Cognitive, pedagogical, and psychological factors will significantly influence it, including AI literacy, AITPCK, and self-efficacy. The paper concludes by sharing recommendations for future research. Through this lens, the research contributes to the literature on AI acceptance in a more comprehensive sense within mathematics teacher education. The findings also indicate that for PSMTs to use these AI tools in a meaningful and effective way, they must be provided not only with technical knowledge but also with learning activities that give these technologies pedagogical value, the development of pedagogical integration competencies, and the strengthening of self-efficacy beliefs about their use. Therefore, teacher education programs need to provide structured learning opportunities that integrate AI-related knowledge, pedagogical application, and hands-on experience.

Footnotes

ORCID iD

Nazan Mersin

Ethical Considerations

This study was approved by the Social and Human Sciences Ethics Committee of Istanbul Medeniyet University (Approval No. E-38510,686-050.04-2500085275).

Consent for Publication

The author confirms that they consent to the publication of this manuscript.

Author Contribution

The author is responsible for all parts of the study, including conceptualization, data collection, analysis, and manuscript preparation.

Funding

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

Declaration of Conflicting Interests

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

Data Availability Statement

The data that support the findings of this study are not publicly available due to privacy and ethical restrictions but are available from the corresponding author upon reasonable request.*

Author Biography

Nazan Mersin is a faculty member in mathematics education at Istanbul Medeniyet University. Her research focuses on teachers’ use of technology in education, technology integration in middle school mathematics teaching, mathematical connections, problem posing, and the history of mathematics. Her academic work contributes to the development of effective teaching practices in mathematics education.

Appendix

References

Aksekili

Kan

(2024). Öğretmenlerin eğitimde yapay zekâ kullanımına yönelik tutum ölçeği geliştirme: Geçerlik ve güvenirlik çalışması. 21. Yüzyılda Eğitim ve Toplum, 13(39), 525–542. https://izlik.org/JA59HR58RF.

Al Abdullatif

A. M.

(2024). Modeling teachers’ acceptance of generative artificial intelligence use in higher education: The role of AI literacy, intelligent TPACK, and perceived trust. Education Sciences, 14(11), 1209. https://doi.org/10.3390/educsci14111209

Al Adwan

A. S.

Al-Adwan

Abbasi

G. A.

Albelbisi

N. A.

Habibi

(2023). Extending the technology acceptance model (TAM) to predict university students’ intentions to use metaverse-based learning platforms. Education and Information Technologies, 28(11), 15381–15413. https://doi.org/10.1007/s10639-023-11816-3

Al Darayseh

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

Al Darayseh

Mersin

(2025). Integrating generative AI into STEM education: Insights from science and mathematics teachers. International Electronic Journal of Mathematics Education, 20(3), 1–13. https://doi.org/10.29333/iejme/16232

Aldemir

Bicer

Kilinc

Moon

Kwok

(2025). Exploring emergent AI-TPACK competencies in a two-week AI literacy module for preservice teachers. Teaching and Teacher Education, 168, 105231. https://doi.org/10.1016/j.tate.2025.105231

Chai

C. S.

Zhou

Shen

Zheng

Chen

(2023). Modeling English teachers’ behavioral intention to use artificial intelligence in middle schools. Education and Information Technologies, 28(5), 5187–5208. https://doi.org/10.1007/s10639-022-11286-z

Anderson

J. C.

Gerbing

D. W.

(1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological bulletin, 103(3), 411–423. https://doi.org/10.1037/0033-2909.103.3.411

Awang

L. A.

Yusop

F. D.

Danaee

(2025). Current practices and future direction of artificial intelligence in mathematics education: A systematic review. International Electronic Journal of Mathematics Education, 20(2), em0823. https://doi.org/10.29333/iejme/16006

10.

Ayanwale

M. A.

Sanusi

I. T.

Adelana

O. P.

Aruleba

K. D.

Oyelere

S. S.

(2022). Teachers’ readiness and intention to teach artificial intelligence in schools. Computers and Education: Artificial Intelligence, 3, 100099. https://doi.org/10.1016/j.caeai.2022.100099

11.

Bandura

(1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84(2), 191–215. https://doi.org/10.1037/0033-295X.84.2.191

12.

Bandura

(1986). Social foundations of thought and action: A social cognitive theory. Prentice Hall.

13.

Bandura

(1997). Self-efficacy: The exercise of control. W. H. Freeman.

14.

Baydaş

Göktaş

(2017). A model for preservice teachers’ intentions to use ICT in future lessons. Interactive Learning Environments, 25(7), 930–945. https://doi.org/10.1080/10494820.2016.123227

15.

Bećirović

Polz

Tinkel

(2025). Exploring students’ AI literacy and its effects on their AI output quality, self-efficacy, and academic performance. Smart Learning Environments, 12(1), 29. https://doi.org/10.1186/s40561-025-00384-3

16.

Biton

Segal

(2025). Learning to craft and critically evaluate prompts: The role of generative AI (ChatGPT) in enhancing pre-service mathematics teachers’ TPACK and problem-posing skills. International Journal of Education in Mathematics, Science and Technology, 13(1), 202–223. https://doi.org/10.46328/ijemst.4654

17.

Canbazoğlu Bilici

Tanrısevdi

Yıldız Durak

J. Çakıroğlu.

2024. Yapay Zeka Teknolojik Pedagojik Alan Bilgisi (YZ-TPAB) Ölçeğinin Türkçeye Uyarlaması . 5th International Conference on Engineering and Applied Natural Sciences. August 25-26, 2024, Konya/Turkey.

18.

Caratiquit

L. J.

(2022). Influence of technical support on technology acceptance model to examine the project PAIR E-Learning system in distance learning modality. Participatory Educational Research, 9(5), 468–485. https://doi.org/10.17275/per.22.124.9.5

19.

Carolus

Koch

M. J.

Straka

Latoschik

M. E.

Wienrich

(2023). MAILS-meta AI literacy scale: Development and testing of an AI literacy questionnaire based on well-founded competency models and psychological change and meta-competencies. Computers in Human Behavior: Artificial Humans, 1(2), 100014. https://doi.org/10.1016/j.chbah.2023.100014

20.

Çelebi

Yılmaz

Demir

Karakuş

(2023). Artificial intelligence literacy: An adaptation study. Instructional Technology and Lifelong Learning, 4(2), 291–306. https://doi.org/10.52911/itall.1401740

21.

Celik

(2023). Towards Intelligent-TPACK: An empirical study on teachers’ professional knowledge to ethically integrate artificial intelligence (AI)-based tools into education. Computers in Human Behavior, 138, 107468. https://doi.org/10.1016/j.chb.2022.107468

22.

Chassignol

Khoroshavin

Klimova

Bilyatdinova

(2018). Artificial intelligence trends in education: A narrative overview. Procedia Computer Science, 136, 16–24. https://doi.org/10.1016/j.procs.2018.08.233

23.

Chen

Lin

(2020). Artificial intelligence in education: A review. IEEE Access, 8, 75264–75278. https://doi.org/10.1109/ACCESS.2020.2988510

24.

Chin

Chew

C. M.

Suseelan

(2022). Mathematics learning from concrete to abstract (1968–2021): A bibliometric analysis. Participatory Educational Research, 9(4), 445–468. https://doi.org/10.17275/per.22.99.9.4

25.

Chin

W. W.

(1998). The partial least squares approach to structural equation modeling. In Marcoulides

G. A.

(Ed.), Modern methods for business research (pp. 295–336). Lawrence Erlbaum Associates.

26.

Chiu

T. K. F.

(2022). Applying the self-determination theory (SDT). To explain student engagement and motivation in online learning during the COVID-19 pandemic. Journal of Research on Technology in Education, 54(S1), s14–s30. https://doi.org/10.1080/15391523.2021.1891998

27.

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

28.

Chou

C.-M.

Shen

T.-C.

Shen

T.-C.

Shen

C.-H.

(2024). Developing and validating an AI-supported teaching applications’ self-efficacy scale. Research and Practice in Technology Enhanced Learning, 19, 035. https://doi.org/10.58459/rptel.2024.19035

29.

Daher

(2025). Integrating AI literacy into teacher education: A critical perspective paper. Discover Artificial Intelligence, 5(1), 217. https://doi.org/10.1007/s44163-025-00475-7

30.

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

31.

Dermeval

Paiva

Bittencourt

I. I.

Vassileva

Borges

(2018). Authoring tools for designing intelligent tutoring systems: A systematic review of the literature. International Journal of Artificial Intelligence in Education, 28(3), 336–384. https://doi.org/10.1007/s40593-017-0157-9

32.

Sun

Jiang

Islam

A. Y. M.

(2024). Exploring the effects of AI literacy in teacher learning: An empirical study. Humanities and social sciences communications, 11(1), 1–10. https://doi.org/10.1057/s41599-024-03101-6

33.

Eickelmann

Vennemann

(2017). Teachers’ attitudes and beliefs regarding ICT in teaching and learning in European countries. European Educational Research Journal, 16(6), 733–761. https://doi.org/10.1177/1474904117725899

34.

European Commission . (2023). A European approach to artificial intelligence. Shaping Europe’s Digital Future. https://digital-strategy.ec.europa.eu/en/policies/artificial-intelligence.

35.

Falebita

O. S.

Kok

P. J.

(2025). Artificial intelligence tools usage: A structural equation modeling of undergraduates’ technological readiness, self-efficacy and attitudes. Journal for STEM Education Research, 8(2), 257–282. https://doi.org/10.1007/s41979-024-00132-1

36.

Falk

C. F.

Biesanz

J. C.

(2015). Inference and interval estimation methods for indirect effects with latent variable models. Structural Equation Modeling: A Multidisciplinary Journal, 22(1), 24–38. https://doi.org/10.1080/10705511.2014.935266

37.

Fornell

Larcker

D. F.

(1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104

38.

Franke

Sarstedt

(2019). Heuristics versus statistics in discriminant validity testing: a comparison of four procedures. Internet research, 29(3), 430–447. https://doi.org/10.1108/IntR-12-2017-0515

39.

Gadanidis

(2017). Artificial intelligence, computational thinking, and mathematics education. The International Journal of Information and Learning Technology, 34(2), 133–139. https://doi.org/10.1108/IJILT-09-2016-0048

40.

Gado

Kempen

Lingelbach

Bipp

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

41.

Getenet

(2024). Pre-Service Teachers and ChatGPT in Multistrategy Problem-Solving: Implications for Mathematics Teaching in Primary Schools. International Electronic Journal of Mathematics Education, 19(1), 1–12. https://doi.org/10.29333/iejme/14141.

42.

Gravemeijer

K. P. E.

(2005). What makes mathematics so difficult, and what can we do about it? In Santos

Canavarro

A. P.

Brocardo

(Eds.), Educação matemática: Caminhos e encruzilhadas (pp. 83–101). APM.

43.

Guo

Shi

Zhai

(2025). Developing and validating an instrument for teachers’ acceptance of artificial intelligence in education. Education and Information Technologies, 30(10), 13439–13461. https://doi.org/10.1007/s10639-025-13338-6

44.

Gurer

M. D.

Akkaya

(2022). The influence of pedagogical beliefs on technology acceptance: A structural equation modeling study of pre-service mathematics teachers. Journal of Mathematics Teacher Education, 25(4), 479–495. https://doi.org/10.1007/s10857-021-09504-5

45.

Hair

J. F.

Black

W. C.

Babin

B. J.

Anderson

R. E.

(2019). Multivariate data analysis (8th ed.). Cengage.

46.

Hava

Babayiğit

Ö.

(2025). Exploring the relationship between teachers’ competencies in AI-TPACK and digital proficiency. Education and Information Technologies, 30(3), 3491–3508. https://doi.org/10.1007/s10639-024-12939-x

47.

Henseler

Ringle

C. M.

Sarstedt

(2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115–135. https://doi.org/10.1007/s11747-014-0403-8

48.

Hong

J. C.

Hwang

M. Y.

Tai

K. H.

, et al. (2019). Improving cognitive certitude with calibration mediated by cognitive anxiety, online learning self-efficacy and interest in learning Chinese. pronunciation. Educational Technology Research and Development, 67(3), 597–615. https://doi.org/10.1007/s11423-018-9628-4

49.

Hsu

(2016). Examining EFL teachers’ technological pedagogical content knowledge and the adoption of mobile-assisted language learning: A partial least square approach. Computer Assisted Language Learning, 29(8), 1287–1297. https://doi.org/10.1080/09588221.2016.1278024

50.

L. T.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. https://doi.org/10.1080/10705519909540118

51.

Ibili

Resnyansky

Billinghurst

(2019). Applying the technology acceptance model to understand maths teachers’ perceptions towards an augmented reality tutoring system. Education and Information Technologies, 24(5), 2653–2675. https://doi.org/10.1007/s10639-019-09925-z

52.

Jang

(2024). AI literacy and intention to use text-based GenAI for learning: The case of business students in Korea. Informatics, 11(3), 54. https://doi.org/10.3390/informatics11030054

53.

Zhong

Lyu

Niu

Zhan

(2025). How does AI literacy affect individual innovative behavior: The mediating role of psychological need satisfaction, creative self-efficacy, and self-regulated learning. Education and Information Technologies, 30(11), 1–30. https://doi.org/10.1007/s10639-025-13437-4

54.

Joo

Y. J.

Park

Lim

(2018). Factors influencing preservice teachers’ intention to use technology: TPACK, teacher self-efficacy, and technology acceptance model. Journal of Educational Technology & Society, 21(3), 48–59. http://www.jstor.org/stable/26458506.

55.

Kadıoğlu-Akbulut

Çetin-Dindar

Acar-Şeşen

Küçük

(2023). Predicting preservice science teachers’ TPACK through ICT usage. Education and Information Technologies, 28(9), 11269–11289. https://doi.org/10.1007/s10639-023-11657-0

56.

Kandlhofer

Steinbauer

Hirschmugl-Gaisch

Huber

(2016). Artificial intelligence and computer science in education: From kindergarten to university. In 2016 IEEE frontiers in education conference (FIE) (pp. 1–9). IEEE. https://doi.org/10.1109/FIE.2016.7757570

57.

Khong

Celik

T. T.

Lai

V. T. T.

Nguyen

Bui

(2023). Examining teachers’ behavioural intention for online teaching after COVID-19 pandemic: A large-scale survey. Education and Information Technologies, 28(5), 5999–6026. https://doi.org/10.1007/s10639-022-11417-6

58.

Kline

(2015). A handbook of test construction (psychology revivals): Introduction to psychometric design. Routledge.

59.

Kline

R. B.

(2023). Principles and practice of structural equation modeling (5th ed.). Guilford Press.

60.

Koh

J. H. L.

Chai

C. S.

Tsai

C. C.

(2013). Examining practicing teachers’ perceptions of technological pedagogical content knowledge (TPACK) pathways: A structural equation modeling approach. Instructional Science, 41(4), 793–809. https://doi.org/10.1007/s11251-012-9249-y

61.

Kong

S. C.

Cheung

W. M. Y.

Zhang

(2022). Evaluating artificial intelligence literacy courses for fostering conceptual learning, literacy and empowerment in university students: Refocusing to conceptual building. Computers in Human Behavior Reports, 7, 100223. https://doi.org/10.1016/j.chbr.2022.100223

62.

Kong

S. C.

Yang

Hou

(2024). Examining teachers’ behavioural intention of using generative artificial intelligence tools for teaching and learning based on the extended technology acceptance model. Computers and Education: Artificial Intelligence, 7, 100328. https://doi.org/10.1016/j.caeai.2024.100328

63.

Kwon

Ottenbreit-Leftwich

A. T.

Sari

A. R.

Khlaif

Zhu

Nadir

Gok

(2019). Teachers’ self-efficacy matters: Exploring the integration of mobile computing device in middle schools. TechTrends, 63(6), 682–692. https://doi.org/10.1007/s11528-019-00402-5

64.

Laupichler

M. C.

Aster

Schirch

Raupach

(2022). Artificial intelligence literacy in higher and adult education: A scoping literature review. Computers and Education: Artificial Intelligence, 3, 100101. https://doi.org/10.1016/j.caeai.2022.100101

65.

Lee

Ali

Zhang

DiPaola

Breazeal

(2021). Developing middle school students' AI literacy. In Proceedings of the 52nd ACM technical symposium on computer science education (pp. 191–197). ACM.

66.

(2024). Assessing Chinese primary mathematics teachers’ self-efficacy for technology integration: Development and validation of a multifaceted scale. Asian Journal for Mathematics Education, 3(2), 231–253. https://doi.org/10.1177/27527263241254496.

67.

(2025). Integrating artificial intelligence in primary mathematics education: Investigating internal and external influences on teacher adoption. International Journal of Science and Mathematics Education, 23(5), 1283–1308. https://doi.org/10.1007/s10763-024-10515-w

68.

Manzari

(2025). AI utilization in primary mathematics education: a case study from a southwestern Chinese city. Education and Information Technologies, 30(9), 11717–11750. https://doi.org/10.1007/s10639-025-13315-z

69.

Nugraha

M. G.

(2025). Validating a context-specific TPACK scale for primary mathematics education in China. International Electronic Journal of Mathematics Education, 20(2). https://doi.org/10.29333/iejme/15934

70.

Lim

E. M.

(2023). The effects of pre-service early childhood teachers’ digital literacy and self-efficacy on their perception of AI education for young children. Education and Information Technologies, 28(10), 12969–12995. https://doi.org/10.1007/s10639-023-11724-6

71.

Lin

Van Brummelen

(2021). Engaging teachers to co-design integrated AI curriculum for K-12 classrooms. In Proceedings of the 2021 CHI conference on human factors in computing systems (pp. 1–12). ACM.

72.

Long

Magerko

(2020). What is AI literacy? Competencies and design considerations. In In Proceedings of the 2020 CHI conference on human factors in computing systems (pp. 1–16). ACM.

73.

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

74.

MacCallum

R. C.

Browne

M. W.

Sugawara

H. M.

(1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1(2), 130–149. https://doi.org/10.1037/1082-989x.1.2.130

75.

MacKinnon

D. P.

Lockwood

C. M.

Williams

(2004). Confidence limits for the indirect effect: Distribution of the product and resampling methods. Multivariate Behavioral Research, 39(1), 99–128. https://doi.org/10.1207/s15327906mbr3901_4

76.

Mailizar

Burg

Maulina

(2021). Examining university students’ behavioural intention to use e-learning during the COVID-19 pandemic: An extended TAM model. Education and Information Technologies, 26(6), 7057–7077. https://doi.org/10.1007/s10639-021-10557-5

77.

Mangarin

R. A.

Caballes

D. O.

(2024). Difficulties in learning mathematics: A systematic review. International Journal of Research and Scientific Innovation, 11(9), 401–405. https://doi.org/10.51244/ijrsi.2024.1109037

78.

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

79.

Miranda

F. J.

Chamorro-Mera

(2025). Exploring the adoption of generative artificial intelligence tools among university teachers. Higher Education Research and Development, 45(3), 1–17. https://doi.org/10.1080/07294360.2025.2559648

80.

Mishra

Koehler

M. J.

(2006). Technological pedagogical content knowledge: A framework for teacher knowledge. Teachers College Record, 108(6), 1017–1054. https://doi.org/10.1111/j.1467-9620.2006.00684.x

81.

Mishra

Warr

Islam

(2023). TPACK in the age of ChatGPT and generative AI. Journal of Digital Learning in Teacher Education, 39(4), 235–251. https://doi.org/10.1080/21532974.2023.2247480

82.

Mohamed

M. Z. b.

Hidayat

Suhaizi

N. N. b.

Sabri

N. b. M.

Mahmud

M. K. H. b.

Baharuddin

S. N. b.

(2022). Artificial intelligence in mathematics education: A systematic literature review. International Electronic Journal of Mathematics Education, 17(3), em0694. https://doi.org/10.29333/iejme/12132

83.

Nardi

Ranieri

(2019). Comparing paper‐based and electronic multiple‐choice examinations with personal devices: Impact on students’ performance, self‐efficacy and satisfaction. British Journal of Educational Technology, 50(3), 1495–1506. https://doi.org/10.1111/bjet.12644

84.

D. T. K.

Leung

J. K. L.

Chu

S. K. W.

Qiao

M. S.

(2021). Conceptualizing AI literacy: An exploratory review. Computers and Education: Artificial Intelligence, 2, 100041. https://doi.org/10.1016/j.caeai.2021.100041

85.

D. T. K.

Leung

J. K. L.

M. J.

Yim

I. H. Y.

Qiao

M. S.

Chu

S. K. W.

(2022). AI education and AI literacy. In AI literacy in K–16 classrooms (pp. 9–19). Springer. https://doi.org/10.1007/978-3-031-18880-0_2

86.

Nikolopoulou

Gialamas

Lavidas

Komis

(2021). Teachers’ readiness to adopt mobile learning in classrooms: A study in Greece. Technology, Knowledge and Learning, 26(1), 53–77. https://doi.org/10.1007/s10758-020-09453-7

87.

Ning

Zhang

Zhou

Wijaya

T. T.

(2024). Teachers’ AI-TPACK: Exploring the relationship between knowledge elements. Sustainability, 16(3), 978. https://doi.org/10.3390/su16030978

88.

Pal

Patra

(2021). University students’ perception of video-based learning in times of COVID-19: A TAM/TTF perspective. International Journal of Human-Computer Interaction, 37(10), 903–921. https://doi.org/10.1080/10447318.2020.1848164

89.

Pan

Luo

(2025). Incorporating AI literacy and AI anxiety into TAM: Unraveling Chinese scholars’ behavioral intentions toward adopting AI-assisted literature reading. IEEE Access, 13, 38952–38963. https://doi.org/10.1109/ACCESS.2025.3546572

90.

Phan

H. P.

Ngu

B. H.

Yeung

A. S.

(2017). Achieving optimal best: Instructional efficiency and the use of cognitive load theory in mathematical problem solving. Educational Psychology Review, 29(4), 667–692. https://doi.org/10.1007/s10648-016-9373-3

91.

Podsakoff

P. M.

MacKenzie

S. B.

Lee

J. Y.

Podsakoff

N. P.

(2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879–903. https://doi.org/10.1037/0021-9010.88.5.879

92.

Preacher

K. J.

Selig

J. P.

(2012). Advantages of monte carlo confidence intervals for indirect effects. Communication Methods and Measures, 6(2), 77–98. https://doi.org/10.1080/19312458.2012.679848

93.

Yunhong

Changping

(2025). Bridging AI literacy and UTAUT constructs: Structural equation modeling of AI adoption among Chinese university students. Humanities and Social Sciences Communications, 12(1), 1–13. https://doi.org/10.1057/s41599-025-05775-y

94.

Runge

Hebibi

Lazarides

(2025). Acceptance of pre-service teachers towards artificial intelligence (AI): The role of AI-related teacher training courses and AI-TPACK within the technology acceptance model. Education Sciences, 15(2), 167. https://doi.org/10.3390/educsci15020167

95.

Russell

Norvig

(2003). Artificial Intelligence: A Modern Approach. Pearson Education.

96.

Santos-Trigo

(2024). Problem solving in mathematics education: Tracing its foundations and current research-practice trends. ZDM–Mathematics Education, 56(2), 211–222. https://doi.org/10.1007/s11858-024-01578-8

97.

Schiavo

Businaro

Zancanaro

(2024). Comprehension, apprehension, and acceptance: Understanding the influence of literacy and anxiety on acceptance of artificial intelligence. Technology in Society, 77, 102537. https://doi.org/10.1016/j.techsoc.2024.102537

98.

Schreiber

J. B.

Stage

F. K.

King

Nora

Barlow

E. A.

(2006). Reporting structural equation modeling and confirmatory factor analysis results: A review. The Journal of Educational Research, 99(6), 323–338. https://doi.org/10.3200/JOER.99.6.323-338

99.

Schwarzer

Jerusalem

(1995). Generalized self-efficacy scale. In: Weinman

Wright

Johnston

(Eds.), Measures in health psychology: A user’s portfolio. Causal and control beliefs (pp. 35–37). NFER-NELSON.

100.

Selwyn

(2022). Education and technology: Key issues and debates (3rd ed.). Bloomsbury Publishing.

101.

Shin

(2022). Teaching mathematics integrating intelligent tutoring systems: Investigating prospective teachers’ concerns and TPACK. International Journal of Science and Mathematics Education, 20(8), 1659–1676. https://doi.org/10.1007/s10763-021-10221-x

102.

Shirvani

(2014). Pre-service teachers’ attitudes toward using technology in schools. The Journal of Literacy and Technology, 15(1), 33–53. https://literacyandtechnology.org/wp-content/uploads/2023/09/jlt_15_1.pdf.

103.

Şimşek

A. S.

Cengiz

G. Ş. T.

Bal

(2025). Extending the TAM framework: Exploring learning motivation and agility in educational adoption of generative AI. Education and Information Technologies, 30(15), 20913–20942. https://doi.org/10.1007/s10639-025-13591-9

104.

Sofyan

Habibi

Sofwan

Yaakob

M. F. M.

Alqahtani

T. M.

Jamila

Wijaya

T. T.

(2023). TPACK–UotI: The validation of an assessment instrument for elementary school teachers. Humanities and Social Sciences Communications, 10(1), 1–7. https://doi.org/10.1057/s41599-023-01533-0

105.

Son

(2024). Intelligent tutoring systems in mathematics education: A systematic literature review using the substitution, augmentation, modification, redefinition model. Computers, 13(10), 270. https://doi.org/10.3390/computers13100270

106.

Yang

(2024). AI literacy curriculum and its relation to children’s perceptions of robots and attitudes towards engineering and science: An intervention study in early childhood education. Journal of Computer Assisted Learning, 40(1), 241–253. https://doi.org/10.1111/jcal.12867

107.

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

108.

Teo

(2011). Factors influencing teachers’ intention to use technology: Model development and test. Computers & Education, 57(4), 2432–2440. https://doi.org/10.1016/j.compedu.2011.06.008

109.

Teo

(2012). Examining the intention to use technology among pre-service teachers: An integration of the technology acceptance model and theory of planned behavior. Interactive Learning Environments, 20(1), 3–18. https://doi.org/10.1080/10494821003714632

110.

Teo

Lee

C. B.

Chai

C. S.

Wong

S. L.

(2009). Assessing the intention to use technology among pre-service teachers in Singapore and Malaysia: A multigroup invariance analysis of the technology acceptance model (TAM). Computers & Education, 53(3), 1000–1009. https://doi.org/10.1016/j.compedu.2009.05.017

111.

Tondeur

Scherer

Baran

Siddiq

Valtonen

Sointu

(2019). Teacher educators as gatekeepers: Preparing the next generation of teachers for technology integration in education. British journal of educational technology, 50(3), 1189–1209. https://doi.org/10.1111/bjet.12748

112.

Tram

N. H. M.

(2025). Unveiling the drivers of AI integration among language teachers: Integrating UTAUT and AI-TPACK. Computers in the Schools, 42(2), 100–120. https://doi.org/10.1080/07380569.2024.2441155

113.

Tschannen-Moran

Hoy

A. W.

(2001). Teacher efficacy: Capturing an elusive construct. Teaching and Teacher Education, 17(7), 783–805. https://doi.org/10.1016/S0742-051X(01)00036-1

114.

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

115.

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

116.

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

117.

Wang

Rau

P. L. P.

Yuan

(2023). Measuring user competence in using artificial intelligence: Validity and reliability of artificial intelligence literacy scale. Behaviour & Information Technology, 42(9), 1324–1337. https://doi.org/10.1080/0144929X.2022.2072768

118.

Wakhata

Mutarutinya

Balimuttajjo

(2023). Exploring the impact of Stein et al.’s levels of cognitive demand in supporting students’ mathematics heuristic problem-solving abilities. In Frontiers in Education, 8. https://doi.org/10.3389/feduc.2023.949988

119.

Wang

Cui

Yuan

(2025). Artificial intelligence in higher education: The impact of need satisfaction on artificial intelligence literacy mediated by self-regulated learning strategies. Behavioral Sciences, 15(2), 165. https://doi.org/10.3390/bs15020165

120.

Wardat

Tashtoush

M. A.

AlAli

Jarrah

A. M.

(2023). ChatGPT: A revolutionary tool for teaching and learning mathematics. Eurasia Journal of Mathematics, Science and Technology Education, 19(7), em2286. https://doi.org/10.29333/ejmste/13272

121.

T. T.

Lee

H. Y.

Wang

W. S.

Lin

C. J.

Huang

Y. M.

(2023). Leveraging computer vision for adaptive learning in STEM education: Effect of engagement and self-efficacy. International Journal of Educational Technology in Higher Education, 20(1), 53. https://doi.org/10.1186/s41239-023-00422-5

122.

Gao

Trainin

(2025). Developing an AI-TPACK framework: Exploring the mediating role of AI attitudes in pre-service TCSL teachers’ self-efficacy and AI-TPACK. Education and Information Technologies, 30(15), 1–25. https://doi.org/10.1007/s10639-025-13630-5

123.

Yang

(2022). Artificial intelligence education for young children: Why, what, and how in curriculum design and implementation. Computers and Education: Artificial Intelligence, 3, 100061. https://doi.org/10.1016/j.caeai.2022.100061

124.

Yang

(2025). Multidimensional analysis of art education teachers’ attitudes and self-efficacy toward artificial intelligence: Exploring relationships and strategies for enhancement. Interactive Learning Environments, 33(4), 2824–2847. https://doi.org/10.1080/10494820.2024.2423175

125.

Yildiz Durak

(2018). Digital story design activities used for teaching programming effect on learning of programming concepts, programming self‐efficacy, and participation and analysis of student experiences. Journal of Computer Assisted Learning, 34(6), 740–752. https://doi.org/10.1111/jcal.12281

126.

Yoon

Hwang

Lee

Roh

K. H.

Kwon

O. N.

(2024). Students’ use of generative artificial intelligence for proving mathematical statements. ZDM - Mathematics Education, 56(7), 1531–1551. https://doi.org/10.1007/s11858-024-01629-0

127.

Zhang

Aslan

A. B.

(2021). AI technologies for education: Recent research & future directions. Computers and education: Artificial Intelligence, 2, 100025. https://doi.org/10.1016/j.caeai.2021.100025

128.

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

129.

Zhao

Zhang

(2025). Understanding teachers’ adoption of AI technologies: An empirical study from Chinese middle schools. Systems, 13(4), 302. https://doi.org/10.3390/systems13040302