Structural Modeling of Cognitive Load Reduction in ChatGPT-Supported Learning

The increasing integration of generative artificial intelligence into higher education raises important questions about its role in supporting learning and cognitive processes. This study examines the relationships between ChatGPT use and cognitive load reduction, focusing on the mediating role of user technological confidence and the moderating role of affective trust within the framework of cognitive load theory. Using secondary survey data from 12,085 higher education students across multiple countries, the proposed model was examined using Partial Least Squares Structural Equation Modeling (PLS-SEM). The findings indicate that user technological confidence plays a dual role: it is a significant direct predictor of cognitive load reduction and also mediates the relationship between ChatGPT use and cognitive load reduction, its direct association. The structural model explained 40.2% of the variance in CLR, and respondent-level and country-cluster bootstrap intervals supported the stability of the principal associations. A particularly noteworthy finding is that affective trust exhibits negative moderating effects, weakening the positive relationships between ChatGPT use and cognitive load reduction, as well as between technological confidence and cognitive load reduction. These findings highlight that the role of generative artificial intelligence in cognitive load reduction should be considered not only in relation to technology use itself but also in the context of users’ individual psychological characteristics. Construct operationalization was strengthened through item-level mapping, independent nomological checks, and content-aligned sensitivity analysis. The primary PLS model retained all 12,085 cases (R2 = 0.402 for CLR); the indirect association was 0.085 (95% CI 0.078–0.092), and both interaction effects were negative but small (f2 < 0.002). The direction of the model remained stable when narrower, content-aligned indicators were used.