From AI assistance to pedagogical reflection: a rubric-mediated model for generative AI in teacher education

The integration of Generative Artificial Intelligence (GenAI) into teacher education necessitates a transition from technical proficiency to critical pedagogical judgment. This study adopts a Design-Based Research (DBR) approach to evaluate the impact of an analytical mediation model based on cognitive oversight rubrics. Utilizing a sample of pre-service teachers ( N = 195), the study analyzed the evolution of perceived self-efficacy and the quality of didactic artifacts designed using Large Language Models (LLMs). Resultsindicate significant improvement across all dimensions, notably the strengthening of pedagogical judgment ( t = 35.12; p < .001) with a large effect size ( d > 0.80). A pivotal finding is the emergence of the self-efficacy paradox: while actual design competence and quantitative self-efficacy significantly increased, qualitative evidence demonstrates that this confidence became more cautious, reflective, and critically calibrated, indicating a shift toward conscious, self-regulated competence. However, persistent variability in STEAM complexity suggests that technological scaffolding encounters a performance ceiling when lacking a solid foundation of Pedagogical Content Knowledge (PCK). The study concludes that human-in-the-loop design and critical mediation are essential to mitigate cognitive offloading and ensure teacher epistemic agency in the algorithmic era.