Stage-specific generative AI use and dependency in learning: associations with test performance, perceived learning, and learning satisfaction

Generative artificial intelligence (GAI) is increasingly used in education, but its role across learning stages remains unclear. This study examined whether GAI use during learning comprehension, task-solving, or task-reviewing was associated with test performance, perceived learning, and learning satisfaction. Using a randomized between-subjects experimental design, 94 participants were recruited and 88 were included in the final analyses; prior knowledge and gender were controlled as covariates. In the regression model, the task-solving group scored higher on the main test than the learning-comprehension group. However, greater self-reported GAI dependency was associated with lower test performance in the task-solving group, while it was associated with higher learning satisfaction and perceived learning in the task-solving and task-reviewing groups. These findings suggest a divergence between objective and subjective learning outcomes under higher GAI reliance. The results highlight the need for structured and critical GAI use and for future research using stronger designs and process-level measures.