AIGC affordance and student self-regulation in private undergraduate education: a serial mediation model

Synthesis: A pragmatic mixed-methods study of 689 undergraduates in industry-education integration programs at a private Chinese university, plus interviews with 8 instructors, tested how the affordances of AI-generated content (AIGC) tools translate into self-regulated learning. Using partial least squares structural equation modeling, the authors found that perceived AIGC affordance raised AIGC Self-Efficacy (beta = 0.583) and self-regulated learning directly (beta = 0.316), and that self-efficacy in turn lifted learning motivation (beta = 0.565) and self-AI Regulation in Education (beta = 0.250). Feedback quality strongly predicted satisfaction (beta = 0.712) but did not improve self-efficacy or self-regulation, and satisfaction had no significant effect on self-regulation. The headline claim is that AIGC works as a psychological catalyst rather than a productivity tool: benefits run through confidence and motivation, not through the tool itself.

Key Findings

1. Perceived AIGC affordance significantly increased AIGC self-efficacy (beta = 0.583, p < 0.001) and self-regulated learning (beta = 0.316, p < 0.05), supporting the affordance-actualisation premise of the model.
2. Self-efficacy was the pivotal mediator: it predicted learning motivation (beta = 0.565) and self-regulated learning (beta = 0.250), and it significantly mediated the path from affordance to motivation (indirect effect 0.329) and from affordance to self-regulation (0.145).
3. Motivation carried the strongest weight on self-regulation (beta = 0.527, p < 0.01), and the serial path affordance to self-efficacy to motivation to self-regulation was significant (0.173).
4. The quality of AIGC assessment feedback predicted satisfaction strongly (beta = 0.712) but not self-efficacy (beta = 0.131, n.s.); satisfaction in turn had no significant effect on self-regulated learning (beta = 0.032, n.s.).
5. Qualitative interviews documented systemic implementation gaps in the industry-education programs: curriculum-industry misalignment, limited project diversity and weak enterprise engagement.

Why satisfaction is not self-regulation

The non-significant paths are the analytical core of the paper. Good AI feedback made students happier without making them better self-regulators, and feeling satisfied did not translate into planning, monitoring or adjusting their own learning. The authors read this as a warning against evaluating AIGC integration by satisfaction metrics alone: positive affect is a weak proxy for the cognitive and motivational mechanisms that actually drive self-regulation. Their proposed design implication is to target activities at building students' confidence with the tool and at making the motivational payoff of using it explicit, rather than assuming that a pleasant, well-functioning assistant will produce deep learning.

Measurement quality was high by conventional standards — factor loadings 0.891 to 0.933, Cronbach's alpha 0.895 to 0.936, composite reliability 0.925 to 0.954, average variance extracted 0.813 to 0.848, SRMR 0.058 for the estimated model, and 10,000 bootstrap resamples. The design remains cross-sectional and single-institution, however, so the directional mediation claims rest on theory rather than on temporal ordering.

What this means for practice

• Instructors. Structure AIGC tasks so students direct the tool — deciding what to delegate and what to work through themselves — rather than receiving finished output, because perceived affordance worked through AIGC self-efficacy (beta = 0.583, p < 0.001), the strongest lever on motivation and self-regulation in this model.
• Instructors. Embed prompt-and-reflect routines and verification rubrics that require students to justify and triangulate AI output, since self-efficacy (beta = 0.565 on motivation) and motivation (beta = 0.527 on self-regulation) carried the regulatory path while good feedback did not.
• Designers. Stop treating well-received AI feedback as evidence of learning gains: feedback quality predicted satisfaction (beta = 0.712) but not self-efficacy (beta = 0.131, n.s.), and satisfaction did not predict self-regulated learning (beta = 0.032, n.s.), so build for confidence and motivation and evaluate against regulation measures instead of satisfaction scores.
• Designers. Fix program-level conditions alongside tool-level design — project variety, industry alignment and enterprise involvement — because the instructor interviews show the tool operating inside weak program structures, and the serial path only activates with authentic institutional support.
• Researchers. Treat the serial order as a chain to be tested rather than assumed, collecting repeated measures of affordance, self-efficacy, motivation and self-regulation so the sequence can be examined rather than inferred from a cross-sectional survey.

Limitations

• The quantitative data come from one cross-sectional survey collected in October–November 2024 in a single private undergraduate institution in Guangzhou: 1,000 questionnaires distributed through the institution's LMS produced 689 valid responses (68.9%), and the authors state that cross-sectional data preclude strong causal claims about the serial mediation.
• Every construct was self-reported on seven-point Likert items, so the affordance, self-efficacy, motivation and self-regulation paths share method variance.
• Reverse causality remains plausible on the authors' own account — high self-regulated learners may selectively perceive AIGC affordances, and self-efficacy and motivation may reciprocally reinforce each other.
• The qualitative strand consists of 8 instructors (enterprise mentors and academic supervisors, interviews of roughly 30 minutes) rather than students, and it documents curriculum–industry misalignment, limited project diversity and weak enterprise engagement rather than student-level mechanisms.

Citation

Liang, S.-Z., Li, Z.-C., Hsu, Y.-M., & Xu, J.-L. (2026). AIGC affordance and student self-regulation in private undergraduate education: a serial mediation model of self-efficacy and learning motivation. Frontiers in Psychology, 17, 1800950.