Can incorporating GenAI chatbots into L2 writing instruction promote student agency and writing performance? Insights from a mixed-methods quasi-experimental study

We conducted a mixed-methods quasi-experimental research design to examine the effects of incorporating GenAI into L2 writing instruction on Chinese university English as a foreign language (EFL) learners’ agency and writing performance over the writing course of an academic semester. Participants comprised 116 first-year Chinese EFL undergraduate students recruited from two intact classes at a public university in southern China, and they were assigned to experimental and control groups. During the eight-week intervention, students in the experimental group (n = 59) were engaged in writing instruction supported by GenAI tools, through which they used prompts to generate feedback on their argumentative writing, and revised three argumentative writing essays based on this feedback. Students in the control group (n = 57) completed the same writing tasks independently but relied solely on self-correction. All participants completed a standardized national writing test and a learner agency questionnaire at three time points: before the intervention, immediately after the intervention, and four weeks after the intervention. We used a linear mixed-effects modeling approach to analyze quantitative data and employed a thematic analysis to analyze qualitative data from reflective journals and semi-structured interviews. Overall, the findings show a varied pattern of effects rather than consistently positive. While GenAI-supported instruction enhanced specific dimensions of learner agency (self-cognition and responsible action), several other dimensions presented minimal or non-significant improvements over time. These results indicate that GenAI may strengthen learners’ awareness and sense of responsibility, but its influence does not extend to promoting more advanced self-regulation or enduring autonomous learning behaviors. Meanwhile, incorporating GenAI chatbots into writing instruction contributes to substantial improvements in writing performance. The qualitative findings further disclose intricate patterns in how GenAI reshapes the writing process, fostering deeper cognitive engagement and strategic regulation, while also introducing tensions associated with authorship, identity, and cognitive offloading. Overall, the findings reveal that GenAI does not primarily function as a direct amplifier of learner agency, but rather as a mediational means that reconfigures how agency is enacted in L2 writing.