When vendor reputation does not matter: Trust decoupling and pragmatic trust in Generative AI adoption among university students.

Abstract

Trust Transfer Theory predicts that trust in a reputable developer extends to its products, yet this assumption has rarely been tested for non-deterministic generative AI (GenAI) systems. This study examines how trust forms and transfers across conceptual (AI in general), institutional (vendor), and product (specific GenAI tool) levels, and how these trust layers relate to students' adoption intention. Survey data from 480 undergraduate students who actively use GenAI at a public research university in Indonesia were analysed using partial least squares structural equation modeling (PLS-SEM), complemented by importance-performance map analysis and necessary condition analysis. Eight of ten hypotheses were supported. Attitude toward AI and social influence predicted trust in AI, which in turn predicted both vendor trust and trust in the specific GenAI tool. Trust in the tool was also predicted by problem-solving capability and user interface quality, and it predicted adoption intention (R2 = 0.32). Two paths were not supported. AI anxiety was not associated with trust in AI (β = -0.02, p = 0.564), a result that applies to students who have already adopted GenAI. Vendor trust did not predict trust in the tool once the other antecedents were accounted for (β = 0.03, p = 0.450). We interpret this pattern as trust decoupling, in which product-level trust rests on direct performance evidence, termed pragmatic trust, rather than on vendor reputation. The findings suggest a boundary condition for trust transfer in GenAI contexts and inform how higher education institutions evaluate and integrate GenAI tools.