International Journal of
Information and Education Technology

Abstract

This study integrates the Elaboration Likelihood Model (ELM) and the Technology Acceptance Model (TAM) to examine how trust mediates the cognitive and emotional mechanisms underlying students’ continued use of generative Artificial Intelligence (AI) chatbots—specifically, ChatGPT. A bibliometric analysis of 21 studies from the Web of Science database (2010–2025) reveals limited integration of ELM and TAM in educational AI contexts, highlighting a theoretical gap. To address this, we developed a dual-path model combining cognitive constructs (e.g., perceived usefulness, information quality) and affective cues (e.g., fairness, accountability, transparency), with trust serving as a mediating variable. A survey of 359 Vietnamese university students was analyzed using Structural Equation Modeling (SEM) and Multi-Group Analysis (MGA) to examine how ChatGPT usage duration moderates the persuasive process. Findings indicate that peripheral cues (e.g., fairness, accountability) significantly shape trust across user groups. However, the central route (information quality) on trust is heavily moderated by user experience: while it exerts a strong effect on low-frequency users, its impact is significantly attenuated—to less than half the strength—among high-frequency users. Furthermore, trust strongly predicts both satisfaction and perceived usefulness, with satisfaction emerging as the most powerful determinant of continuance usage intention. The study contributes to theory by validating the integration of ELM and TAM in the context of AI adoption and highlighting the moderating role of usage frequency. Practical implications include promoting AI literacy and designing chatbot interfaces with greater transparency to meet the cognitive expectations of experienced users.