Key points are not available for this paper at this time.
Purpose This study examines the continuance usage intention of ChatGPT among Indonesian university students. Considering the limitations of traditional TAM constructs for generative AI technology, this study developed an integrated generative-AI response adoption framework (GRAF) with two new constructs: AI relevant performance and AI comfort expectancy. Design/methodology/approach This study employs a quantitative approach with purposive sampling. Data were collected from 688 university students who used ChatGPT in higher-education institutions on the island of Java, Indonesia. Two new constructs – AI relevant performance and AI comfort expectancy – were developed through four stages of validation and tested using exploratory factor analysis. Hypothesis testing was conducted using partial least squares structural equation modeling (PLS-SEM). Findings The findings reveal that information quality and system quality do not directly affect continuance usage intention but are fully mediated by AI-relevant performance and AI comfort expectancy. Satisfaction is the strongest predictor of continuance intention. However, satisfaction did not mediate the relationship between system quality and continuance intention, indicating that ChatGPT users developed a direct evaluation pathway through AI-specific factors. Research limitations/implications This study did not include the factors of ethical concerns and privacy-related impacts, nor did it differentiate adoption patterns across disciplines. Theoretically, this research broadens the understanding that generative AI technology requires a more contextual evaluation construct than the traditional TAM. Practical implications Higher education institutions must integrate AI relevance assessment into curriculum design, provide contextual prompt engineering training, and create safe spaces for AI exploration to enhance users’ psychological comfort. Originality/value This study contributes to the literature by developing two new constructs (AI relevant performance and AI comfort expectancy) that are empirically validated, as well as identifying a full mediation mechanism, which indicates that the evaluation of generative AI technology requires a different cognitive pathway than conventional technology.
Triangga et al. (Fri,) studied this question.