GenAI literacy and GenAI-assisted self-regulated learning behaviors: the roles of learning agency and challenge emotions

The rapid integration of generative artificial intelligence (GenAI) into higher education has sparked critical conversations surrounding academic autonomy, cognitive offloading, and effective self-regulated learning. As universities worldwide navigate the shifting digital paradigm, a new multi-institutional study conducted in China has shed light on the cognitive and psychological mechanisms that dictate how future educators interact with AI-driven tools. Published in Frontiers in Psychology, the empirical research investigates the intricate connections between GenAI literacy and self-regulated learning behaviors among special education undergraduates, highlighting the pivotal mediating roles of learning agency and challenge emotions.

Background Context and Research Framework

The proliferation of large language models and advanced AI systems has dramatically restructured how students acquire information, synthesize complex concepts, and receive academic feedback. While these technologies promise unprecedented personalization—particularly in specialized fields like special education, where adaptive interventions and intelligent assessments are paramount—they also introduce substantial risks. Educators and researchers have expressed growing concern that uncritical or excessive reliance on GenAI tools can foster cognitive offloading, thereby eroding students’ critical thinking and cognitive autonomy.

To counteract these pitfalls, educational experts emphasize the necessity of self-regulated learning (SRL), an active, goal-directed framework wherein learners monitor and adjust their cognition, motivation, behavior, and emotional states. However, possessing advanced AI tools does not automatically guarantee effective self-regulation. Recognizing this gap, a research team spearheaded by Juan Yang from Shaanxi Normal University sought to explore whether a student’s foundational GenAI literacy—defined as the comprehensive competency to understand, evaluate, and responsibly utilize AI—serves as a catalyst for disciplined, self-directed learning behaviors.

Methodology and Chronology of the Study

The empirical investigation utilized a cross-sectional online survey design distributed across multiple stages between May and June 2025. Leveraging academic networks and institutional collaborations, the research team targeted full-time undergraduate students majoring in special education across universities in four Chinese provinces: Shaanxi, Sichuan, Yunnan, and Hainan. These regions were chosen to capture a diverse geographic and institutional cross-section of teacher-training programs.

Data collection was executed digitally via the Wenjuanxing platform and distributed through regional academic networks on WeChat. Initial outreach yielded 503 completed questionnaires. Through a rigorous screening process designed to eliminate disengaged responses, excessive missing data, and anomalous completion times, the research team retained 434 valid submissions, achieving an effective response rate of 86.3%.

The demographic breakdown of the final participant pool revealed a distinct gender distribution: 396 participants identified as female (91.2%) and 38 as male (8.8%). While heavily skewed toward female students, the authors noted that this ratio accurately reflects the broader demographic composition of special education undergraduate programs in China. In terms of academic progression, first-year students accounted for 17.7%, second-year students for 18.2%, third-year students for 60.1%, and fourth-year seniors for 3.9%.

Evaluation Instruments and Quantitative Findings

The survey instruments deployed established, psychometrically validated scales adapted for GenAI contexts. GenAI literacy was evaluated across four distinct dimensions—knowledge, skills, attitudes/values, and ethics—using a 7-point Likert scale (Cronbach’s α = 0.885). Learning agency was measured across 34 items evaluating key abilities, active actions, and mental characteristics (Cronbach’s α = 0.942). Challenge emotions, which encompass positive affective states such as flow, excitement, and playfulness, were captured using a specialized four-item scale (Cronbach’s α = 0.865). Finally, GenAI-assisted self-regulated learning behaviors (SRLB) were quantified using an 11-item adapted schedule (Cronbach’s α = 0.911).

Through robust statistical modeling utilizing the R software environment and lavaan package, the researchers tested a hypothesized serial mediation model while controlling for gender and year of study. Bivariate correlations initially confirmed that GenAI literacy, learning agency, and challenge emotions were all positively and significantly associated with SRLB.

The core regression and bootstrapping analyses yielded several critical insights:

• Direct Association: GenAI literacy exhibited a statistically significant direct positive association with GenAI-assisted self-regulated learning behaviors (B = 0.219, p = 0.006), validating Hypothesis 1.
• Individual Mediation: Learning agency proved to be a robust individual mediator, accounting for a substantial indirect effect between literacy and regulatory behaviors (B = 0.345, 95% CI [0.234, 0.461]). Challenge emotions also demonstrated a statistically significant, albeit smaller, individual mediating effect (B = 0.061, 95% CI [0.001, 0.126]).
• Serial Mediation: Crucially, the analysis confirmed the hypothesized serial pathway. GenAI literacy sequentially influenced learning agency, which subsequently shaped challenge emotions, ultimately driving enhanced self-regulated learning behaviors (B = 0.075, 95% CI [0.030, 0.122]).

Implications for Special Education and Teacher Training

The implications of these findings extend far beyond general higher education, carrying profound weight for the specialized training of future special educators. As artificial intelligence becomes deeply embedded in classrooms serving students with diverse developmental and learning needs, future educators must possess more than basic technical proficiency. They must cultivate active agency and manage their emotional appraisals when confronting complex technological tools.

The study’s authors emphasize that curriculum developers must restructure AI literacy training programs. Rather than focusing solely on prompt engineering or software operation, teacher-preparation courses should intentionally foster student agency—encouraging undergraduates to critically interrogate AI-generated lesson plans, diagnostic assessments, and adaptive interventions rather than accepting them at face value. Furthermore, addressing the affective dimension—specifically nurturing positive challenge emotions rather than anxiety or passive reliance—is paramount for maintaining cognitive engagement.

Institutional Responses and Future Outlook

While the findings offer a vital empirical roadmap for integrating generative AI into teacher education, the researchers acknowledged certain methodological boundaries. The reliance on self-report questionnaires, despite stringent statistical controls against common method variance, indicates a need for future longitudinal and experimental designs to firmly establish causality. Additionally, the geographic concentration within specific provinces and the heavily female-dominated sample highlight the necessity for broader, cross-cultural, and gender-balanced replications in subsequent research cycles.

Financial support for the underlying research was provided by the 2025 Shaanxi Higher Vocational Education Teaching Reform Research Project, alongside institutional grants dedicated to advancing pedagogical reform in special education. As academic institutions worldwide grapple with the dual promises and perils of the artificial intelligence era, this study underscores that technological competency must be inextricably linked with human agency and emotional self-regulation to truly empower the next generation of educators.