The impact of artificial intelligence literacy on learning engagement among university students: the mediating role of psychological capital and the moderating role of professional identity

Synthesis: Wang (2026) tests a moderated mediation model on 1,198 undergraduates from four universities in Zhengzhou, China, showing that AI literacy not only directly predicts learning engagement but also works indirectly by building psychological capital — a resource-transformation chain in which a technological cognitive resource becomes a psychological resource and then a behavioral one. Drawing on Conservation of Resources Theory with Self-Determination Theory and Broaden-and-Build as complements, the study finds that professional commitment does not itself drive engagement but acts as a contextual amplifier, roughly doubling the efficiency with which psychological capital converts into engagement. The finding reframes AI literacy training as necessary but insufficient: its educational payoff depends on simultaneous psychological and identity work.

Key Findings

• AI literacy directly predicts learning engagement. Using partial least squares structural equation modeling, the study finds that students who can understand, apply, and evaluate AI show substantially higher engagement, and the model explains a large share of the variance in engagement. This supports the baseline hypothesis.
• Psychological capital partially mediates the literacy–engagement link. AI literacy builds the psychological resources of self-efficacy, hope, optimism, and resilience, and those resources in turn predict engagement. Roughly half of the literacy effect runs through this psychological pathway, and the direct effect remains significant on top of it — so this is partial, not full, mediation.
• Professional commitment is a pure moderator, not a predictor. Commitment to one's major does not itself raise engagement (its direct effect is negligible and non-significant). Instead, it conditions the strength of the psychological capital → engagement relationship, so its role is an amplifier rather than an independent cause.
• The effect nearly doubles at high commitment. Simple slope analysis shows that psychological capital converts into engagement roughly twice as efficiently among students high in professional commitment as among those low in it — the study's core boundary-condition claim.
• The indirect effect strengthens with commitment. The path from AI literacy through psychological capital to engagement is stronger at high professional commitment, and the index of moderated mediation is significant, confirming the whole "conditional" chain rather than just the final link.
• Demographic differences were substantial. Males scored higher on AI literacy while females scored higher on learning engagement; student leaders reported higher engagement, psychological capital, and commitment; and freshmen and seniors showed relatively higher engagement and commitment than middle-year students.
• Professional commitment is empirically independent of the other constructs. Its correlations with both AI literacy and learning engagement are near zero, which the author reads as evidence that it works as a clean boundary condition rather than a confounded proxy predictor.

Study Design & Method

A cross-sectional questionnaire survey of undergraduate students at four universities in Zhengzhou, Henan Province, conducted February–April 2026. A stratified random sampling design was used, with strata formed from the full combination of gender, grade level, and student leadership status, with proportional allocation by stratum and mixed-mode administration (online via Wenjuanxing and offline paper). After excluding invalid responses — excessive missing data, patterned responding, or unrealistically fast completion — 1,198 valid questionnaires remained. The sample's gender balance (about half female) closely matched national higher-education enrollment figures.

Instruments: AI literacy via a six-dimension scale (perception, comprehension, knowledge, skills, evaluation, innovation; 22 items); professional commitment via a four-dimension scale (ideal, affective, continuance, normative; 27 items); psychological capital via the four-dimension Positive Psychological Capital Questionnaire covering self-efficacy, resilience, hope, and optimism (27 items); and learning engagement via the Utrecht Work Engagement Scale for Students measuring vigor, dedication, and absorption (17 items). Factor structure, reliability, and discriminant validity were all adequately supported, and procedural remedies for common method bias included temporal separation of predictors and outcomes, differential scale anchors across constructs, and anonymity assurances.

Analysis used SPSS for descriptives and correlations and SmartPLS for partial least squares structural equation modeling, chosen for its prediction orientation and its handling of a complex model combining mediation, moderation, a second-order construct, and an interaction term. Mediation and moderated mediation were tested by bootstrap resampling with bias-corrected confidence intervals, with simple slopes plotted at one standard deviation above and below the mean of the moderator.

What this means for practice

• Instructors. Pair AI Literacy coursework with psychological-capital building — Self-Efficacy, hope, optimism, and resilience — through scaffolded, progressively harder AI-assisted projects, because roughly half of the literacy effect on engagement ran through psychological capital (R² = 0.507 for psychological capital, 0.497 for engagement).
• Designers. Front-load identity work for students with weak professional commitment — professional orientation courses, alumni career storytelling, industry practice, and mentoring that ties content to careers — since commitment nearly doubled the conversion of psychological capital into engagement (conditional effect 0.374 at high commitment versus 0.204 at low).
• Instructors. Differentiate by profile: students high in AI literacy but low in psychological capital benefit most from small-step, achievable group projects, while students low on both dimensions need foundational support on both at once.
• Administrators. Screen psychological capital and professional commitment inside academic monitoring to flag the two priority groups, and remember that commitment is a moderator rather than a driver — its direct effect on engagement was negligible.
• Researchers. Treat professional commitment as a contextual amplifier rather than a direct predictor; that repositioning reconciles previously inconsistent findings that AI use sometimes enhances and sometimes erodes active learning behavior.

Limitations

• The cross-sectional design precludes strict causal inference; reverse or reciprocal paths (e.g. engaged students seeking out AI literacy) cannot be ruled out. Longitudinal designs or pretest–posttest intervention studies are needed.
• The sample is confined to undergraduates at four universities in Zhengzhou, limiting generalizability across regions, disciplines, and educational levels.
• Only psychological capital as a composite construct was tested as a mediator; future work should disaggregate AI self-efficacy, technological resilience, and other specific mediators.
• Rapid iteration of generative AI is shifting use from shallow instrumental application toward deep collaborative co-creation, so the conceptualization and measurement of AI literacy — and assessment tools capable of capturing adaptability and reflection across usage patterns — must evolve accordingly.

Citation

Wang, N. (2026). The impact of artificial intelligence literacy on learning engagement among university students: the mediating role of psychological capital and the moderating role of professional identity. Frontiers in Psychology, 17, 1892204.