Trust calibration and perceived developmental gains in university students' collaboration with generative AI: an exploratory sequential mixed methods study

Abstract

Research on generative artificial intelligence (GenAI) in higher education has expanded rapidly, but the field still explains student adoption more clearly than it explains how AI supported activity becomes higher quality learning. Prior studies have tended to emphasize use intention, broad AI literacy, ethical concern, or general perceptions of impact, while offering less precise accounts of the process linking supportive educational contexts to students' trust judgments, regulatory behavior, and developmental outcomes. To address this gap, the present study proposes and tests a contextual support, trust calibration, collaborative regulation, and perceived transfer gains model using an exploratory sequential mixed methods design. In Study 1, semi structured interviews with university students were analyzed through open, axial, and selective coding to identify the core dimensions of the model and to inform item development. In Study 2, a 22 item instrument was pilot tested and then administered to 642 students for confirmatory factor analysis and structural equation modeling. The structural model showed that contextual support positively predicted trust calibration and collaborative regulation; trust calibration positively predicted collaborative regulation and perceived transfer gains; and collaborative regulation showed the strongest association with perceived transfer gains. The direct path from contextual support to perceived transfer gains was not statistically significant, whereas the chained indirect pathway through trust calibration and collaborative regulation was significant. The findings suggest that developmental benefits in students' collaboration with GenAI are better understood as theory consistent structural relationships shaped by contextual guidance, calibrated trust, and active regulation rather than as automatic consequences of frequent AI use. The study contributes a more differentiated process model for GenAI supported learning in higher education and clarifies the interpretive limits of subjective cross sectional outcome measures.