How generative AI becomes embedded in design learning: a 14-instructional-day intensive longitudinal study of reciprocal dynamics between AI use and self-efficacy

Abstract

Generative AI is increasingly embedded in design education, yet little is known about how students’ daily engagement with generative AI is dynamically associated with their learning-related capability beliefs across successive instructional days. Drawing on self-efficacy theory, this study examined the reciprocal next-instructional-day associations between daily generative AI use in design learning and daily design-learning self-efficacy, and further investigated whether baseline hassles were associated with individual differences in these dynamics. A 14-instructional-day diary study was conducted with 260 Chinese undergraduate design students, yielding 3,558 observed person-day assessments out of 3,640 possible assessments. Daily generative AI use and design-learning self-efficacy were assessed throughout the diary period, whereas hassles were measured at baseline. The data were analyzed using bivariate dynamic structural equation modeling within a Bayesian estimation framework. The results showed significant autoregressive effects for daily generative AI use ( β = 0.198, 95% CrI [0.154, 0.237]) and daily design-learning self-efficacy ( β = 0.138, 95% CrI [0.100, 0.175]). At the within-person level, higher-than-usual generative AI use on one instructional day positively predicted design-learning self-efficacy on the subsequent instructional day ( β = 0.043, 95% CrI [0.006, 0.077]), while higher-than-usual design-learning self-efficacy also positively predicted generative AI use on the subsequent instructional day ( β = 0.059, 95% CrI [0.023, 0.095]). At the between-person level, greater baseline hassles were associated with higher average generative AI use ( β = 0.135), lower average design-learning self-efficacy ( β = −0.254), and a stronger cross-lagged association from generative AI use to subsequent self-efficacy ( β = 0.458). Gender did not significantly predict the main random effects. These findings indicate modest reciprocal next-instructional-day associations between generative AI use and design-learning self-efficacy, suggesting that the two may reinforce one another across instructional days. The stronger association from generative AI use to subsequent self-efficacy among students reporting greater baseline hassles further suggests that contextual demands may shape how generative AI becomes integrated into students’ day-to-day design learning.