When Generative AI Helps and Hurts: Augmentation-Substitution Taxonomy and Recovery Pattern in University Students’ Experiences

Generative AI is now widely used in higher education, yet the same tool can produce positive and negative outcomes for the same user, and the reasons for this remain underexplored. Using technology affordance theory, this study develops an empirically grounded taxonomy of interaction mechanisms based on students’ retrospective accounts. The 71 participants, who began university after the public release of generative AI, completed surveys and reported 254 experiences through interviews and written reflections. Three key findings emerged: first, most students reported both positive and negative experiences, with no measured person-level trait robustly associated with experiential valence; second, 17 interaction mechanisms were identified, organized by whether AI complemented or replaced cognitive tasks; third, some students shifted from negative to more positive trajectories. Experiential valence thus appears more closely associated with interaction context than with person-level traits, although the evidence is retrospective rather than causal.