Personalisation and Trust Formation in Generative AI Interaction: A Structural Equation Modelling Study of Doubao users’ Continuance Intention

Abstract

With the growing adoption of generative AI in personal and business practices, users’ sustained use relies on system performance and trust judgments during long-term interactions. Unlike the traditional static personalization, generative AI can adjust its answers through multi-rounds interaction and demonstrates perceived intelligence, adaptivity and self-learning techniques. As a result, dynamic personalization cues might be a significant contributor in the development of human-machine trust. Based on both the Stimulus-Organism-Response (SOR) model and the Ability-Benevolence-Integrity (ABI) trust model, this research investigates how perceived AI intelligence and perceived AI self-learning ability impact users’ continuance intention via perceived ability, perceived benevolence, and trust. With data from Chinese generative AI users by a structured questionnaire, this study applies covariance-based structural equation modeling (CB-SEM) to test measurement model, structural model and mediating effects. The results show that perceived AI intelligence significantly positively affects perceived ability. Perceived AI self-learning ability positively affects both perceived ability and perceived benevolence. Further, perceived ability and perceived benevolence contribute to trust, and trust substantially strengthens continuance intention. The mediation results demonstrate that perceived ability, perceived benevolence and trust have significant mediating effects on the relationship between personalization ability cues and continuance intention. By distinguishing perceived intelligence from perceived self-learning ability, this study contributes to a better understanding of human-machine trust building in the dynamic interaction context of new generative AI era.