Push or Pull? Understanding switching intentions from human services to GAI agents through the PPM framework

With the rapid advancement of artificial intelligence, generative AI agents such as ChatGPT are transforming knowledge acquisition and production, raising critical questions about why and how users switch from traditional human services to AI-driven systems. This study applies the Push–Pull–Mooring (PPM) framework to examine the determinants of users' switching intentions from traditional human services to generative AI agents. Data were collected from 451 valid survey responses across five countries (United States, United Kingdom, Singapore, China, and Pakistan). Using Partial Least Squares Structural Equation Modeling (PLS-SEM), the study assessed both first- and second-order constructs. The results showed that functional, emotional, and social dissatisfaction significantly increased users’ switching intentions, whereas perceived inefficiency did not exert a significant effect. On the pull side, attractiveness—driven by perceived intelligence, ease of use, and emotional capability—emerged as the strongest predictor of switching. Notably, emotional capability exerted a stronger influence than functional factors, underscoring the importance of affective design in GAI agents. Among mooring factors, inertia shaped by switching costs and habitual dependence negatively affected switching, while subjective norms positively moderated adoption by amplifying social influence. Theoretically, this study extended the PPM framework to AI-mediated knowledge services by identifying emotional capability as a novel pull factor in technology adoption. Practically, the findings highlighted the need to improve empathetic responsiveness, reduce switching costs through user-friendly onboarding, and tailor strategies to regional disparities in digital literacy. Together, these insights provided a cross-cultural perspective on user migration from human services to GAI agents, advancing understanding of human–machine interaction in the intelligent era. © 2025 Elsevier Ltd.