Most of us in the AI-in-education space have built our advocacy around literacy. Teach students to understand AI, the thinking goes, and use will follow. A new study out of Türkiye complicates that assumption in a way I can’t ignore. Türk et al. (2025) tested a chain model of student AI acceptance, and the link turned out to be the weak one.
The researchers surveyed 356 university students, running their general attitude toward AI through two mediators, AI literacy and AI self-efficacy, with AI learning anxiety added as a moderator. The model is tidy and grounded in the Technology Acceptance Model. The result is the part worth slowing down for.
Attitude Does Most of the Work
The single strongest finding is how much a positive attitude predicts acceptance on its own. Türk et al. found that attitude directly drove acceptance, and that direct effect held firm even after both mediators entered the model. The full mediation chain explained only about 11% of attitude’s total effect. Put plainly, the feeling came first and carried most of the weight.
That tracks with what I’ve watched in classrooms. Students who feel good about AI reach for it. Students who feel wary keep their distance, no matter how much they know. The authors are clear that “having a positive attitude towards the benefits of artificial Intelligence will increase the acceptance of artificial Intelligence in the educational process” (p. 8). Attitude isn’t a soft variable here. It’s the engine.
The Literacy Surprise
Here’s where the study earns its keep. Of the two mediators, self-efficacy did the heavy lifting and literacy came through weak. Stranger still, Türk et al. report a negative correlation between AI literacy and self-efficacy. Students who knew more about AI tended to feel less confident using it.
The authors offer a reading I find genuinely unsettling: “students who possess high levels of AI literacy may experience low self-efficacy because they recognize the significant potential and function of AI” (p. 9). The more clearly you see how vast and fast-moving AI is, the smaller your own grip on it can feel.
They take the idea one step further with a consumer-versus-producer frame. Students who use AI mainly for convenience may start to see themselves as passive users, and that self-image chips away at confidence. It echoes the concern I keep returning to in the work on metacognitive laziness (Fan et al., 2025), where the worry isn’t whether students can access AI but what happens to their sense of agency around it. Literacy without a producer’s mindset can leave students competent and timid at the same time.
Anxiety Changes the Math
The moderator tells its own story. AI learning anxiety reshaped the attitude-acceptance link. The connection was steepest for students with low anxiety and flattened as anxiety climbed. A good attitude still helped anxious students, but it bought them less acceptance than it bought their calmer peers.
I like that the authors resist a tidy villain story here. Anxiety dampened the path, yes, but they note it can also raise extrinsic motivation, so its role isn’t purely negative. That’s the kind of complexity I want to see held, not resolved cheaply. Anxiety is a brake on some students and a strange spur for others.
References
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. _British Journal of Educational Technology, 56 (2), 489–530. https://doi.org/10.1111/bjet.13544
- Türk, N., Batuk, B., Kaya, A., & Yıldırım, O. (2025). What makes university students accept generative artificial intelligence? A moderated mediation model. BMC Psychology, 13, Article 1257. https://doi.org/10.1186/s40359-025-03559-2