What drives the adoption of AI-facilitated automated written corrective feedback tools in L2 writing? A SEM study based on Activity Theory

Despite the widespread use of AI-facilitated automated written corrective feedback (AWCF) tools in second/foreign language (L2) education, existing technology acceptance models have not yet fully incorporated comprehensive external constructs, especially when Activity Theory can guide the selection of socio-environmental, individual, and systemic predictors in Chinese English as a foreign language (EFL) learning contexts. This study adopted a mixed-methods design to examine seven determinants of AWCF acceptance and readiness through a cross-sectional survey using five-point Likert scales. A total of 392 valid questionnaires were collected through convenience sampling from middle-tier Chinese universities. Quantitative data were analyzed through Partial Least Squares Structural Equation Modeling (PLS-SEM) in SmartPLS to examine relationships among the constructs. In addition, 146 open-ended qualitative responses were analyzed through inductive coding, assisted by AntConc, to capture learners’ suggestions and contextualize the statistical path results. The quantitative findings confirmed that perceived ease of use and usefulness significantly and positively influenced AWCF acceptance and readiness, with ease of use also positively associated with usefulness. Among the external predictors, systemic characteristics emerged as the strongest determinant of perceived ease of use (β = 0.517, 95% CI [0.452, 0.577], p < 0.001), while perceived usefulness (β = 0.514, 95% CI [0.418, 0.598], p < 0.001) and perceived ease of use (β = 0.352, 95% CI [0.268, 0.447], p < 0.001) directly predicted students’ AWCF acceptance and readiness. The model explained 62.7% of the variance in AWCF acceptance and readiness, and its Q² value reached 0.354, indicating moderate explanatory and predictive power for the outcome variable. Gender did not significantly moderate any of the eight proposed relationships (interaction βs ≤ |0.097| , 95% CIs crossing zero, p values > 0.05), suggesting model stability across male ( N = 105) and female ( N = 287) subgroups, although the gender imbalance limited the power to detect interaction effects. Qualitative keyword analysis identified two themes (i.e., language revision and software operation) and 17 subthemes concerning learners’ suggestions for AWCF improvement. Overall, this study advances the prediction of AWCF acceptance and readiness in L2 writing by integrating Activity Theory-derived socio-environmental, individual, and systemic dimensions and refining them into five constructs. The findings offer pedagogical implications for teachers and institutions and call for future longitudinal, cross-cultural, and broader-sample designs to establish causal direction and validate the model’s generalizability, particularly regarding the strong effect of systemic characteristics and the absence of gender moderation.