Perceived Desirable Difficulty in AI Tutoring: Engagement, Retention Confidence, and Continued Use

Abstract

This study examines the role of perceived desirable difficulty in AI tutoring systems and explains how productive learning challenge influences learner engagement, productive struggle orientation, intrinsic motivation, knowledge retention confidence and continued use intention. Although AI tutoring is increasingly used in higher education, much of the existing discussion mainly focuses on usefulness, satisfaction and general acceptance. Due to this reason, less attention has been given to whether students accept AI-supported difficulty as a meaningful learning condition. To address this gap, the study developed a structural model by integrating the Desirable Difficulties Framework, Self-Determination Theory and technology continuance logic. A quantitative cross-sectional survey design was applied, and data were collected from 358 university students who had experience using AI tutoring or AI-assisted learning tools. The proposed model was tested through Partial Least Squares Structural Equation Modelling using SmartPLS 4. The findings show that perceived desirable difficulty is the main psychological and learning driver in the model. It significantly influenced learner engagement, productive struggle orientation, intrinsic motivation, knowledge retention confidence and continued use intention. The result also shows that intrinsic motivation, learner engagement and productive struggle orientation significantly improved knowledge retention confidence, which later influenced continued use intention. Moreover, perceived AI adaptiveness significantly moderated the relationship between perceived desirable difficulty and continued use intention. From the above findings, it can be understood that difficulty in AI tutoring is not harmful when it is productive, adaptive and psychologically meaningful. Rather, manageable challenge can support deeper engagement, stronger motivation, better retention confidence and sustained AI tutoring use.