Learners’ longitudinal trajectories of self-efficacy and grit in interactive AI-supported learning environments: implications for STEM education design

The integration of generative artificial intelligence (GAI) into higher education has reshaped interaction, feedback, and reflection in learning environments. Grounded in a Perception–Belief–Action–Outcome framework, this study examined how early-stage perceptions of AI use were associated with technical university students’ self-efficacy, academic grit, and STEM-integrated competence over time. A longitudinal mixed-methods design was implemented in an AI-supported STEM course. Three-wave longitudinal structural equation modeling was used to examine temporal stability, reciprocal relationships, and multi-stage mediation, while qualitative reflective reports and open-ended responses were analyzed to contextualize the quantitative findings. The results showed that early perceptions of AI feedback quality, usability, and cognitive support were positively associated with subsequent self-efficacy development, sustained effort, and STEM-integrated competence. Self-efficacy and academic grit demonstrated reciprocal belief–action dynamics across the semester, suggesting that confidence and perseverance mutually reinforced one another during AI-supported learning. Qualitative findings further indicated that AI-supported feedback and reflection helped students manage uncertainty, sustain engagement, and integrate knowledge across STEM domains. Theoretically, this study advances a longitudinal belief–action process perspective on AI-supported learning. Practically, the findings highlight the value of interactive, feedback-rich, and reflective AI-supported environments for fostering confidence, persistence, and long-term STEM competence.