Understanding the usefulness–risk paradox in AI writing tool adoption: An extended technology acceptance model study in Chinese University L2 writing contexts

Abstract

Generative artificial intelligence (GenAI) is rapidly reshaping higher education, and AI writing tools have become increasingly prominent in second language (L2) writing contexts. However, the mechanisms underlying students’ adoption of these tools, as well as the perceived consequences of their use, remain insufficiently understood. Grounded in an extended Technology Acceptance Model (TAM), this study examines how AI self-efficacy (ASE), multidimensional perceived risk, and behavioral engagement (BE) are associated with Chinese university students’ adoption of AI writing tools in L2 writing-related tasks and their perceived writing-related outcomes. A total of 518 valid questionnaires were collected from undergraduate students with prior experience using AI writing tools for English writing, revision, translation, or other L2 writing-related tasks, and structural equation modeling was employed to test the proposed acceptance–use–outcome framework. The results show that ASE was positively associated with perceived ease of use (PEOU) and perceived usefulness (PU). While PEOU was negatively associated with overall perceived risk (OPR), PU was positively associated with OPR, indicating that students who perceived AI writing tools as useful also tended to report stronger awareness of potential risks. OPR was not significantly associated with attitudes toward AI writing tools. Attitude was strongly associated with behavioral engagement, which in turn was positively associated with both perceived positive and perceived negative writing-related outcomes, with the positive association being substantially stronger. These findings extend TAM in AI-mediated learning by showing that functional value and risk awareness may coexist rather than offset one another in L2 writing contexts. Because the study used cross-sectional self-report data, the findings should be interpreted as associations among perceived constructs rather than evidence of direct causal effects or objective improvement in writing performance. The study also offers practical implications for the responsible integration of AI writing tools in higher education.