AI-supported higher-order thinking and EFL writing quality: A mechanism-focused study

This research investigated whether students' AI-supported higher-order thinking (critical thinking, creativity, metacognition, synthesis) relates to EFL academic writing quality (AWQ) and through what mechanisms. Using blind-rated essays and survey measures from 492 undergraduates at urban Chinese universities, we estimated a comparative structural model with writing self-efficacy (L2WSE) as a mediator and trust in AI feedback as a moderator. The model explained 58% of the variance in AWQ. All four predictors showed statistically significant but modest direct effects (e.g., β_metacognition≈0.22; β_creativity≈0.12), and L2WSE was positively associated with AWQ. Indirect effects via L2WSE were small yet significant, consistent with a motivation-linked pathway. Interaction terms for trust were positive but small (β ≈ 0.06-0.09). Robustness checks compared a second-order specification of AI-supported higher-order thinking against the four first-order predictors and examined possible nonlinearity in trust. Findings should be interpreted cautiously: the design is cross-sectional, effects are modest, and most predictors are self-reported. Even so, the results suggest that when AI tools are used to support metacognitive planning/monitoring, critical evaluation, synthesis, and creative exploration, students' confidence and writing performance may improve. We discuss implications for instruction that emphasize metacognitive routines and calibrated trust in AI feedback, while outlining directions for longitudinal and experimental research.