<p>This study examines the factors influencing graduate students’ adoption of Generative AI as an assistive tool in the context of thesis writing. Based on a comprehensive framework of driving and constraining forces, three sub-studies investigate the effects of thesis writing self-efficacy, Generative AI self-efficacy, time pressure, and trust in AI on the intention to use Generative AI. Study 1 recruited 395 graduate students from Taiwan to validate the factor structure of the measurement scales and test hypotheses using a covariance-based structural equation model (CB-SEM). Study 2 employed a multi-source approach to collect data from 422 graduate students. Findings from both studies consistently indicate that thesis writing self-efficacy functions as a constraining force, suppressing the intention to use Generative AI. At the same time, time pressure reduces thesis writing self-efficacy. In contrast, Generative AI self-efficacy serves as a driving force, facilitating the intention to use Generative AI, and trust in Generative AI enhances Generative AI self-efficacy. Study 3 conducted interviews with 20 graduate students, providing a more nuanced understanding of the relationship between writing competence and usage intention while identifying two additional inhibitory factors. By integrating quantitative and qualitative evidence, this study highlights the dual role of self-efficacy in the context of technology adoption. It offers empirical insights into the application of Generative AI in higher education.</p>