A dual-path trust-risk model of GenAI continuance: Conversational attributes, psychological evaluations, and configurational patterns.

Abstract

The rapid advancement of generative artificial intelligence (GenAI) has transformed human-AI interaction by enabling more efficient, flexible, and conversational forms of user support. To better understand users' continuance intention, this study draws on the post-adoption satisfaction-continuance logic of the expectation-confirmation model (ECM) and combines it with the stimulus-organism-response (SOR) framework and a trust-risk perspective to examine how perceptions of conversational attributes-namely competence perception, interactivity, anthropomorphism, and information accuracy-are associated with continuance intention through trust, perceived risk, and satisfaction. A mixed analytical approach was adopted, combining partial least squares structural equation modeling (PLS-SEM) and fuzzy-set qualitative comparative analysis (fsQCA) using survey data from 452 GenAI users. The PLS-SEM results showed that competence perception, interactivity, anthropomorphism, and information accuracy were positively associated with trust, whereas only information accuracy was negatively associated with perceived risk. Trust was positively associated with satisfaction, while perceived risk was negatively associated with satisfaction. Trust and satisfaction were directly and positively associated with continuance intention. Perceived risk showed no significant direct relationship with continuance intention but was negatively and indirectly associated with it through satisfaction. The fsQCA identified five configurations associated with high continuance intention, highlighting the equifinality of technological and psychological conditions. These findings suggest that supporting sustained GenAI use may require attention not only to system performance and conversational quality but also to calibrated trust, risk governance, and user satisfaction.