Heutagogy and generative AI: an empirical investigation of self-determined learning on deep learning outcomes

Background The pervasive accessibility of Generative AI poses a growing risk of reinforcing the Surface Approach to learning, potentially undermining the development of students’ capability and self-regulated thinking. Objectives This study aimed to empirically evaluate the effectiveness of a Heutagogical Deep Learning (HDL) framework that integrates structured GenAI-based meta-cognitive scaffolding in promoting deep learning and meta-cognitive regulation. Methods A quasi-experimental mixed-methods design was employed. Using the Revised Study Process Questionnaire and the meta-cognitive Awareness Inventory, learning outcomes in the HDL Intervention Group were compared with those in a Traditional Control Group within an applied Data Science module. Quantitative findings were triangulated with qualitative evidence from reflective portfolios and interviews. Results Results indicated that students in the HDL group achieved significantly higher Deep Approach scores and substantially reduced Surface Approach tendencies compared to the control group. Critically, the intervention produced a strong and significant improvement in the Regulation of Cognition subscale of the MAI, supported by qualitative evidence of double-loop reflection triggered by the GenAI critique protocol. Conclusion These findings empirically validate the HDL framework as an effective pedagogical model for ethically guiding GenAI use, transforming it from a shortcut mechanism into a catalyst for capability development and self-regulated deep learning.