Generative AI and EFL students’ perceived research writing competence: A mixed-methods parallel mediation analysis
TL;DR
It is argued that while GenAI acts as a digital support tool, it also risks cognitive passivity that alters students' self-competence perceptions, and higher education institutions should adapt assessment paradigms and implement structured metacognitive training.
Abstract
This study examines how generative artificial intelligence usage behavior (UG) influences the perceived research writing competence (RWC) of EFL university students in Ho Chi Minh City. Utilizing a sequential explanatory mixed-methods design, the inquiry integrates partial least squares structural equation modeling (PLS-SEM) with a thematic analysis of semi-structured interviews. This paper argues that while GenAI acts as a digital support tool, it also risks cognitive passivity that alters students' self-competence perceptions. The results show that the structural model explains 58.9% of the variance in perceived research writing competence (R2 = 0.589). The frequency of GenAI interaction directly supports learners' writing confidence (β = 0.265, p < 0.001), while active academic discourse socialization (AD) serves as a constructive mediating channel (β = 0.133, p < 0.001) that enables students to internalize discipline-specific rhetorical conventions. Conversely, extensive technological interaction induces a state of cognitive dependency (CD) (β = 0.641, p < 0.001), yielding an inflated perceived competence score (β = 0.233, p < 0.001). This dual pathway reveals a paradoxical educational phenomenon whereby algorithmic text-processing efficiency is frequently mistaken for independent scholarly reasoning. Since the indirect effect of CD is larger than that of AD, these outcomes highlight the risk of technological over-reliance. In practical terms, these findings suggest that higher education institutions should adapt assessment paradigms and implement structured metacognitive training.