Adoption intention of generative artificial intelligence among Chinese college students: an extended TAM-UTAUT2 model from the four-helix perspective

To explore the formation mechanism of college students' adoption intention of generative artificial intelligence (GAI)—i.e., the dynamic paths of direct/indirect effects of antecedent variables and interaction effects of moderating variables—this study integrates and extends the traditional Technology Acceptance Model (TAM) and Unified Theory of Acceptance and Use of Technology 2 (UTAUT2). The integration is necessary because traditional TAM focuses on rational cognition, while UTAUT2 lacks the emotional dimension in educational scenarios and the integration of multi-level contexts. A theoretical framework incorporating the “individual-family- institution-region” four-dimensional moderating and collaborative perspective was constructed, and an empirical analysis was conducted using Partial Least Squares Structural Equation Modeling (PLS-SEM). A multi-stage stratified sampling method was adopted, with a sample of 842 college students from five universities in eastern and western China. The scales for UTAUT2 core variables and extended TAM variables in the questionnaire were adapted from previous studies that had undergone reliability and validity verification. Reliability was tested using Cronbach's α and Composite Reliability, while validity was tested using Average Variance Extracted, the Fornell-Larcker criterion, and Heterotrait-Monotrait Ratio. Results showed that the measurement model had acceptable reliability and validity, with good explanatory power (R 2 = 0.743) and predictive validity of the structural model. Specifically, perceived comfort (β = 0.112, p < 0.005), perceived security (β = 0.109, p < 0.05), and emotional dependence (β = 0.497, p < 0.005) all exerted positive effects on perceived usefulness and perceived ease of use. Performance expectancy (β = 0.216, p < 0.005), social influence (β = −0.064, p < 0.05), facilitating conditions (β = 0.143, p < 0.005), and perceived ease of use (β = 0.469, p < 0.005) directly drove adoption intention, whereas the direct effect of perceived usefulness was not significant (β = −0.031, p = 0.523). This result challenges the core assumption of TAM—which emphasizes “priority of rational utility” and confirms the “de- instrumentalization” characteristic of generative AI adoption, meaning user decisions rely more on emotional experience and interactive fluency.Regarding moderating effects: gender negatively moderated the relationship between performance expectancy and adoption intention (β = −0.207, p < 0.05); family structure negatively moderated the relationship between habit and adoption intention (β = −0.228, p < 0.05); university type positively moderated the relationship between performance expectancy and adoption intention (β = 0.251, p < 0.05); and regional differences negatively moderated this relationship (β = −0.251, p < 0.05).In practice, it is suggested that educational authorities strengthen the construction of digital infrastructure in western universities, universities develop differentiated guidance strategies for students majoring in humanities/social sciences and science/engineering, and developers optimize emotional interaction design. This study provides theoretical support for context-adapted strategies for the educational application of generative AI.