Individual Characteristics as Moderators in the AI Literacy–Attitude Relationship: Preservice Math Teachers

Significance of the research

Today, artificial intelligence (AI) applications support decision-making in education, enhance teaching and learning quality, and enable real-time monitoring of student motivation (Arslan, 2020; Gürlek, 2024; Hwang et al., 2020; İşler & Kılıç, 2021; Sharma et al., 2019). Using AI in schools helps create personalized curricula by identifying students’ weaknesses and saves teachers time (Kuprenko, 2020; Luckin, 2017). AI also encourages collaboration, offers instant assessment, and ensures equal opportunities (UNESCO, 2019). Additionally, it supports effective teaching for students with special needs, enables rapid intervention for learning difficulties, reduces teacher workload in crowded classrooms, and helps teachers with content development (Balacheff, 1993; Chassignol et al., 2018; İşler & Kılıç, 2021). However, AI may negatively affect classroom discipline, be vulnerable to system failures and cyber-attacks, and create access issues for socioeconomically disadvantaged students (Turğut et al., 2023). Since artificial intelligence has a close relationship with mathematics, it is important for preservice mathematics teachers—who are the future educators—to have strong literacy and positive attitudes toward AI applications to ensure effective teaching in mathematics classrooms. Moreover, examining preservice mathematics teachers’ AI literacy levels together with their attitudes toward AI is crucial both for the successful integration of AI-based tools in teaching and for shaping the quality of future educational environments. AI literacy involves not only understanding basic concepts and operating systems correctly but also the ability to evaluate ethical consequences and social impacts (Çelebi et al., 2023; Kara, 2024). When preservice mathematics teachers are able to integrate artificial intelligence tools into course design and problem-solving processes, alongside their cognitive and pedagogical knowledge, they can significantly contribute to developing students’ 21st-century skills such as analytical thinking, model building, and data literacy (Kong et al., 2023; Ng et al., 2021). Additionally, their ability to critically assess the limitations and potential biases of AI helps reduce risks at both individual learning and societal levels (Chiu et al., 2020; Dwivedi et al., 2021). Therefore, investigating the AI literacy levels of preservice mathematics teachers supports the safe, fair, and effective use of innovative approaches in education and provides a scientific foundation for designing teacher training curricula and shaping educational policies. However, studies show that even if preservice teachers have positive attitudes toward AI applications, they cannot use AI effectively without foundational knowledge to support these attitudes (Delcker et al., 2024; Frimpong, 2022; Pokrivcakova, 2023). For this reason, instead of investigating AI literacy alone, it is necessary to examine both AI literacy and attitudes towards AI together, as well as other variables that may affect these factors. The literature within the scope of this study indicates that some individual characteristics such as age, gender, and socio-economic status influence attitudes towards artificial intelligence (Gnambs & Appel, 2019; Hajam & Gahir, 2024; Hudson et al., 2017). Moreover, studies show that teachers’ and preservice teachers’ attitudes towards AI are related to their AI literacy levels (Karacan & Çiçek, 2024; Mart & Kaya, 2024; Yim & Wegerif, 2024). However, there are no studies focusing on how individual characteristics of preservice mathematics teachers affect the relationship between their attitudes towards AI and their AI literacy. Therefore, within this study, both the relationship between preservice mathematics teachers’ attitudes and literacy regarding AI and the moderating effects of individual characteristics such as gender, having taken AI-related courses during undergraduate education, daily technology usage time, and interest in AI applications—factors thought to influence this relationship—were examined. In this context, the research problem was organized as: “Is there a significant relationship between preservice mathematics teachers’ AI literacy and attitudes scores, and do these individual characteristics moderate this relationship?” In this frame, the sub-problems were formed as:

• Is there a significant relationship between preservice mathematics teachers’ AI literacy scores and their attitude scores towards AI?

• Does gender have a moderating effect on this relationship?

• Does taking courses on artificial intelligence applications during undergraduate education have a moderating effect on this relationship?

• Does the duration of daily technology use moderate this relationship?

• Does interest in artificial intelligence applications moderate this relationship?

Research Method

This study used a correlational survey model, a quantitative method that reveals how and to what extent variables change together (Fraenkel et al., 2012). In correlational models, a third variable—called a moderator—is included to examine when, under what conditions, and to what degree the relationship between dependent and independent variables holds (Gürbüz, 2019; Hayes & Rockwood, 2017). In this study, pre-service mathematics teachers’ AI literacy scores served as the independent variable, AI attitude scores as the dependent variable, and the moderating effects of individual characteristics —such as gender, taking an AI course during undergraduate education, daily technology use time, and interest in AI applications—were investigated.

Population and Sample

The population of this study includes pre-service teachers enrolled in mathematics education programs at education faculties of state universities in the Central Anatolia Region of Turkey. Using cluster sampling, each of the 13 cities in the region was considered a cluster, and six cities were randomly selected. Data were collected from 856 pre-service teachers studying mathematics education at the education faculties of six state universities in these cities. Participants with missing individual characteristics information were excluded. The distribution of the remaining participants according to the moderating variables is shown in Table 1.

Prior to the administration of the scales, the purpose of the study was explained, and informed consent was obtained from all preservice teachers who volunteered to participate.

Data Collection Tools

The data collection tools are explained below.

Validity and Reliability

Since the study utilized only the total scores from the scales, Cronbach’s α analyses were conducted on these total scores to assess internal consistency. The reliability coefficient was calculated as α = 0.70 for the Artificial Intelligence Literacy Scale and α = 0.82 for the Artificial Intelligence Attitude Scale. While 0.70 is often cited as the lower limit, it is widely recognized as an acceptable threshold for internal consistency in social science research (Hair et al., 2019; Nunnally, 1978). Thus, the instruments were deemed to possess sufficient reliability for the purposes of this study.

For construct validity, confirmatory factor analysis (CFA) was performed on both scales, with model fit indices reported in Table 2.

The fit indices (X²/df ≤ 5, RMSEA ≤ 0.10, AGFI > 0.90, and GFI & CFI > 0.85) were evaluated based on established thresholds (Anderson & Gerbing, 1984; Marsh et al., 1988). As observed in Table 2, both scales met these criteria, indicating an acceptable model fit.

Regarding the data analysis method, the SPSS PROCESS Macro was utilized. Specifically, 5,000 bootstrap resamples were employed. This technique was chosen not to alter the reliability of the scales, but to generate robust bias-corrected confidence intervals for the moderation effects. Bootstrapping is recommended in moderation analyses to make valid inferences about statistical significance without relying on the assumption of a normal sampling distribution (Hayes, 2017).

Coding and Analysis of Data

The Artificial Intelligence Literacy Scale and the Artificial Intelligence Attitude Scale are five-point Likert-type instruments, with item scores ranging from 1 to 5. Accordingly, total scores range from 12 to 60 for the Literacy Scale and from 20 to 100 for the Attitude Scale. Individual characteristics variables were categorically coded and incorporated into the analysis as moderator variables. In this study, preservice mathematics teachers’ AI literacy scores were treated as the independent variable, and their attitudes toward AI as the dependent variable. Gender, taking an AI course during undergraduate education, daily technology use duration, and interest in AI (none/some/high) were examined as moderating variables. A moderating variable affects the strength of the relationship between the independent and dependent variables through interaction, as indicated by a significant change in R² with the inclusion of the interaction term and a 95% confidence interval (BootLLCI–BootULCI) that does not include zero (Aiken et al., 1991; Hayes, 2017; MacKinnon et al., 2004). In this context, moderation analysis was conducted separately for each individual characteristics variable using PROCESS (v4) Model 1 (Hayes, 2022), employing 5,000 bootstrap samples. A conceptual representation of the model is provided in Figure 1.

Figure 1

The Relationship Between AI Literacy Scores and Attitudes Toward AI

Prior to testing moderation effects, the direct relationship between the independent variable—preservice mathematics teachers’ AI literacy scores—and the dependent variable—their attitudes toward AI—was examined. The results revealed a significant positive correlation between AI literacy and attitudes toward AI (r = 0.297, p = 0.00). Although the strength of this relationship approaches a medium level, it was determined that AI literacy scores accounted for approximately 9% of the total variance in attitudes toward AI.

The finding that preservice teachers’ AI literacy scores significantly and positively predict their attitudes toward AI aligns with emerging empirical evidence in the field. Recent studies (Karacan & Çiçek, 2024; Mart & Kaya, 2024; Yim & Wegerif, 2024) have consistently demonstrated that a foundational understanding of AI technologies fosters more positive attitudinal dispositions. Specifically, Yim and Wegerif (2024) argue that as educators gain competence in AI concepts, their apprehension decreases, leading to more favorable attitudes. Most recently, Özden et al. (2025) confirmed this trajectory, demonstrating that increased AI literacy significantly mitigates anxiety and fosters positive attitudes among preservice teachers. However, the current study observed a relatively weak relationship which is corroborative of findings by Özden et al. (2025) and Reyes et al. (2024). These scholars suggest that while literacy is a necessary condition for positive attitudes, it is not solely sufficient; external factors such as ethical concerns and perceived utility also play critical roles. Thus, the modest correlation found in this study reinforces the multifaceted nature of AI adoption, indicating that literacy is a starting point rather than the sole determinant of attitude.

The Moderating Effect of Gender on the Relationship Between AI Literacy Scores and AI Attitude Scores

When the gender variable and its interaction with preservice mathematics teachers’ AI literacy scores were included in the model, the overall model significantly predicted attitudes toward AI (F_{(3, 824)} = 29.78, p < .001). The independent variables accounted for approximately 9.8% of the total variance in AI attitude scores. The results of the moderation analysis for gender are presented in Table 3, while conditional effects and group comparisons are provided in Table 4.

According to Tables 3 and 4, the effect of AI literacy scores on attitudes toward AI was statistically significant and positive for both female (θ_X→Y|(W = 0) = 0.67; p < .001) and male (θ_X→Y|(W = 1) = 0.49; p < .001) preservice teachers. However, the interaction term was negative (–0.18), suggesting a slightly weaker relationship for males. Despite this, the interaction effect was not statistically significant (β = –0.18; se = 0.14; t = –1.26; p = .208). The non-significance of the interaction indicates no statistically significant difference between genders in this relationship; thus, gender does not moderate the association between AI literacy and AI attitudes.

The analysis revealed that gender did not moderate the relationship between AI literacy and attitudes. This aligns with recent data from Diaconu (2025), who found that while gender disparities may exist in raw literacy scores, the predictive relationship between attitudinal dispositions and literacy remains robust across genders. While direct research on this specific moderation is scarce, this nonsignificant finding contributes to the ongoing debate regarding gender gaps in technology adoption. Galindo et al. (2024) similarly reported that digital competence influences attitudes toward AI independently of gender, supporting the notion that the ‘gender digital divide’ may be narrowing in the context of AI literacy. Conversely, Strzelecki and ElArabawy (2024) found that gender’s moderating role is highly context-dependent, being significant in Egyptian samples but not in Polish ones. Our results align with the latter context, suggesting that in certain educational environments, gender may no longer be a primary differentiator in how technical knowledge translates to attitude.

Although some studies (e.g., Guipitacio et al., 2025) found gender to be a significant moderator between self-efficacy and attitudes, others (e.g., Liu, 2025) found no such effect for computer self-efficacy. This discrepancy suggests that moderation dynamics are sensitive to the specific constructs measured (e.g., anxiety vs. literacy) and sample characteristics. The lack of moderation in the current study implies that for preservice mathematics teachers, the cognitive pathway from knowing AI (literacy) to liking AI (attitude) functions similarly across genders, potentially due to the standardized curriculum they undergo, which may homogenize their technological dispositions.

The Moderating Effect of Taking AI Courses During Undergraduate Education on the Relationship Between AI Literacy Scores and AI Attitude Scores

When preservice mathematics teachers’ AI literacy scores, course-taking status, and their interaction were included in the model, the model significantly explained attitudes toward AI (F_{(3, 828)} = 26.66, p < .001). The independent variables accounted for approximately 8.8% of the total variance in attitude scores. Results of the moderation analysis for course-taking status are presented in Table 5, with conditional effects and group comparisons shown in Table 6.

According to Tables 5 and 6, the effect of AI literacy scores on attitudes toward AI was significant and positive for both preservice teachers who took AI courses during their undergraduate education (θ_X→Y|(W = 1) = 0.54; p < .001) and those who did not (θ_X→Y|(W = 0) = 0.55; p < .001). The effects in both groups were very similar, and the small difference was not statistically significant (β = 0.00; se = 0.12; t = –0.01; p = .995). The non-significance of the interaction term indicates that the difference in this relationship between those who took AI courses and those who did not may be coincidental. Consequently, it was concluded that taking AI courses during undergraduate education does not moderate the relationship between preservice mathematics teachers’ AI literacy scores and their attitudes toward AI.

Unexpectedly, taking AI-related courses during undergraduate education did not moderate the relationship between AI literacy and attitudes. This finding diverges from studies suggesting that formal AI training enhances both literacy and positive attitudes (Hong & Kim, 2025; Yue et al., 2025). A plausible explanation for this null effect lies in the nature of the courses and the ubiquity of informal learning. As established by Ertmer and Ottenbreit-Leftwich (2010), technology integration is driven more by practical application than theoretical knowledge; consequently, if the courses taken by participants were predominantly theoretical, their impact on the literacy-attitude link might be negligible. Furthermore, the rapid proliferation of AI tools allows preservice teachers to acquire significant ‘informal knowledge’ through social media and digital platforms (Greenhow & Lewin, 2016), potentially diluting the distinct impact of formal coursework.

While literature emphasizes the necessity of formal AI education (Al-Zyoud, 2020; Chai et al., 2020; Hur, 2024), our findings suggest that merely ‘taking a course’ is insufficient to alter the mechanism between literacy and attitude. Instead, as Wardat et al. (2024) and Sanusi et al. (2024) imply, the quality and pedagogical design of these courses—specifically their focus on hands-on integration strategies—may be more critical than their mere presence in the curriculum.

The Moderating Effect of Duration of Technology Use on the Relationship Between AI Literacy Scores and AI Attitude Scores

When preservice mathematics teachers’ AI literacy scores, daily duration of technology use, and their interaction term were included in the model, they significantly predicted attitudes toward AI (F_{(5, 826)} = 17.53, p < .001). The independent variables explained approximately 9.6% of the total variance in AI attitude scores. The results of the moderation analysis for the duration of technology use are presented in Table 7, with conditional effects and group comparisons shown in Table 8.

Examining Tables 7 and 8, the effect of AI literacy scores on attitudes toward AI was not significant for preservice teachers whose daily technology use was less than 30 minutes (θ_X→Y|W = 0 = –0.15; p = .643). However, this effect was significant for those using technology between 30 minutes and 2 hours (θ_X→Y|(W = 1) = 0.64; p < .001) and for those using it more than 2 hours daily (θ_X→Y|(W = 2) = 0.55; p < .001). In other words, the relationship between AI literacy and attitudes was stronger among preservice teachers who used technology daily for between 30 minutes and 2 hours, as well as those who used it for more than 2 hours. According to Hayes (2022), when the moderator variable is multicategorical, the significance of any relative interaction effect suffices to establish a moderating effect. These findings indicate that daily technology usage duration moderates the relationship between AI literacy and attitudes toward AI.

The analysis established that daily technology usage duration significantly moderates the relationship between AI literacy and attitudes. Specifically, the relationship was nonsignificant for low-usage users (<30 mins) but significant and positive for moderate (30 mins–2 hours) and high-usage (>2 hours) groups. This aligns with Hayes (2022), who posits that significance in any relative interaction effect confirms moderation in multicategorical variables.

While direct comparisons are limited, these results offer a nuanced perspective to Elçiçek’s (2024) findings, which suggested that excessive screen time might correlate with lower literacy. In contrast, our study indicates that increased exposure to technology strengthens the link between literacy and positive attitudes. This discrepancy likely stems from the population difference (general students vs. preservice teachers) and the nature of usage (passive consumption vs. active professional preparation). Essentially, for preservice teachers, higher daily engagement with technology likely provides the necessary experiential context where AI literacy can meaningfully translate into positive professional attitudes.

The Moderating Effect of Interest Level Towards AI on the Relationship Between AI Literacy Scores and AI Attitude Scores

The moderator and interaction terms of preservice mathematics teachers’ AI literacy scores and their interest level in AI applications, when included in the model, significantly explained attitudes toward AI (F_{(5, 826)} = 29.34; p < .001). Furthermore, all independent variables in the model accounted for approximately 15.08% of the total variance in AI attitude scores. The results of the moderation analysis for the interest level variable are presented in Table 9, with conditional effects and group comparisons provided in Table 10.

When the values in Tables 9 and 10 are examined, the effect of AI literacy scores on attitudes towards AI was not statistically significant for preservice teachers who reported no interest in AI applications (θ_X→Y| (W = 0) = –0.19; p = .385). However, the effect was found to be statistically significant and positive for those who reported being somewhat interested (θ_X→Y| (W = 1) = 0.49; p < .001) and very interested (θ_X→Y| (W = 2) = 0.44; p = .001) in AI applications. In other words, AI literacy scores positively influenced attitudes towards AI only among preservice mathematics teachers who expressed some or high levels of interest in AI, whereas no such effect was observed among those who were not interested at all.

Furthermore, the effect of AI literacy on AI attitude scores differed significantly between those who were somewhat interested and those who were not interested (β = 0.68; se = 0.23; t = 2.98; p < .001). A similar significant difference was found between those who were very interested and those who were not interested (β = 0.63; se = 0.25; t = 2.47; p < .001). However, no statistically significant difference was observed between the somewhat interested and very interested groups (β = –0.05; se = 0.15; t = –0.31; p = .756).

The study provides novel empirical evidence that interest in AI moderates the literacy-attitude relationship, albeit in a nonlinear manner. The finding that the relationship is weak or nonexistent for those with ‘no interest’ aligns with Hidi and Renninger (2006) and Schiefele et al. (1992), who argue that interest is a prerequisite for cognitive engagement. Without interest, preservice teachers likely lack the internal motivation to process their AI literacy into a coherent attitudinal stance.

Conversely, the observation that the moderating effect stabilizes between ‘some interest’ and ‘high interest’ groups supports Rotgans and Schmidt’s (2017) threshold hypothesis, suggesting that once a baseline level of interest is ignited, further increases do not proportionally intensify the literacy-attitude link. This finding extends previous work by Kovačević and Demić (2024) and Moon et al. (2024) by demonstrating that while interest drives the acquisition of literacy, its role in converting that literacy into attitude has a saturation point. This highlights the critical need for teacher education programs to focus on sparking initial interest to activate this pathway.