“Brave New World” or not?: A mixed-methods study of the relationship between second language writing learners’ perceptions of ChatGPT, behaviors of using ChatGPT, and writing proficiency

This study investigates the relationship between second or foreign language (L2) writing learners’ perception of ChatGPT, their usage behavior, and L2 writing proficiency, including complexity, accuracy, and fluency. A mixed-methods design was utilized to collect data from 215 university L2 writing learners through surveys and argumentative writing samples, and four students participated in a focus group discussion. Results indicated that performance expectancy and effort expectancy significantly influenced students’ intended behaviors, while the effect of social influence was non-significant. Facilitating conditions significantly predicted actual behaviors. Furthermore, the study found that ChatGPT usage significantly predicted only complexity. During the focus group discussion, students expressed the need for support from teachers. The study highlights the importance of human agency in writing instruction, including clear principles, critical evaluation of ChatGPT feedback, and constant monitoring to reinforce human roles in writing. These findings provide insights into students’ perceptions and behaviors toward AI-based writing tools and inform effective writing instruction in the AI era.