Thoughtless Use of Generative Artificial Intelligence and College Students' Self-Directed Learning: A Multi-Group SEM Analysis of Gender Differences

Synthesis: Zhao & Gu (2026), grounded in Social Cognitive Theory, survey 487 undergraduates in Henan Province, China, and use multi-group SEM to show that the thoughtless use of generative AI (TUGA) — adopting AI answers without critically evaluating or understanding them — significantly undermines self-directed learning (SDL). This negative effect operates both directly and through the partial mediation of self-efficacy (SE) and Motivation. Multi-group analysis reveals substantial gender differences: the negative impact of thoughtless use on motivation was stronger for male students, while effects on self-efficacy were stronger for female students.

Key Findings

• TUGA directly harms self-directed learning. Thoughtless use of GenAI had a significant direct negative effect on undergraduates' SDL (β = −0.42, t = 6.76, p < .001). The model — TUGA, self-efficacy, and motivation — explained 75.3% of the variance in SDL.
• Strongest predictor was motivation. Among direct predictors of SDL, Motivation had the strongest positive effect (β = 0.68, t = 11.63, p < .001), followed by the negative effect of TUGA and the positive effect of Self-Efficacy (β = 0.23, t = 3.93, p < .001).
• TUGA erodes both self-efficacy and motivation. Thoughtless use significantly and negatively predicted motivation (β = −0.54, t = 8.53, p < .001) and self-efficacy (β = −0.37, t = 4.75, p < .001), while self-efficacy positively predicted motivation (β = 0.78, t = 23.56, p < .001).
• Partial mediation through SE and motivation. Indirect effects through motivation (β = −0.37, p < .001), self-efficacy (β = −0.09, p = .012), and the sequential self-efficacy→motivation path (β = −0.20, p = .003) were all significant, alongside the significant direct effect — indicating partial mediation.
• Gender differences. The negative impact of thoughtless use on learning motivation was significantly stronger for male students, while its effects on self-efficacy were significantly stronger for female students (partial measurement invariance: significant variance differences for motivation and SDL, not for SE and TUGA).

Study Design & Method

A quantitative study grounded in Social Cognitive Theory. 487 undergraduates from Henan Province, China, were surveyed via snowball sampling. Structural equation modeling (PLS-SEM with SmartPLS) tested the relationships among thoughtless GenAI use (TUGA), self-efficacy (SE), motivation, and self-directed learning (SDL), including mediation (bootstrapping) and multi-group analysis (PLS-MGA with the Welch–Satterthwaite method) to examine gender differences. Measurement invariance was assessed via MICOM. Model fit was acceptable (SRMR = 0.063).

What this means for practice

• Instructors. Design assignments that pair AI-assisted exploration with independent analysis and reflection, since thoughtless generative AI use had a significant negative effect on self-directed learning (β = −0.42) and the model explained 75.3% of the variance in SDL.
• Instructors. Require students to compare AI outputs against authoritative academic sources and explain their reasoning, targeting the Self-Efficacy path that thoughtless use erodes (β = −0.37) and that in turn feeds Motivation (β = 0.78).
• Administrators. Avoid one-size-fits-all AI guidance: because the multi-group analysis found the negative effect on motivation was stronger for male students while the effect on self-efficacy was stronger for female students, target motivation-oriented support and confidence-building support differently.
• Researchers. Test interventions that rebuild learner Learner Agency and motivation rather than ban or uncritically encourage AI, since motivation was the strongest direct predictor of SDL (β = 0.68).

Limitations

• The 487 undergraduates were surveyed in Henan Province, China, and the sample was 78.4% female, so the gender comparison rests on 105 male respondents and findings are bounded to that context.
• The design is cross-sectional; the authors caution that the bootstrap mediation is statistical, not causal, and that developmental trajectories of thoughtless use cannot be observed.
• All measures are self-report, and only self-efficacy and motivation were modeled as mediators — learning strategies, metacognitive skills, and social support were unmodeled, with motivation treated as a single undifferentiated construct.

Citation

Zhao, H., & Gu, H. (2026). Thoughtless use of generative artificial intelligence and college students' self-directed learning: a multi-group SEM analysis of gender differences. Scientific Reports, 16, 24567.