Résumé:
This study aimed to investigate how EFL students engage with AI-generated feedback from two widely used AI tools, ChatGPT and DeepSeek, during a writing revision task.
Specifically, it aims to compare engagement behaviours across the two tools. The study also explores students’ perceptions of their engagement. A sequential explanatory mixed-methods approach was adopted, combining quantitative data from a quasi-experiment and a post-questionnaire with qualitative insights from a focus group interview. Fifty two Algerian EFL university second year students were randomly assigned to either the ChatGPT or DeepSeek group. Participants wrote a short paragraph on a given topic and then revised it using their assigned AI tool during a controlled laboratory session. Behavioural engagement was measured through the frequency of acceptance, rejection, and modification behaviours. The post-experiment questionnaire, completed by all participants, captured immediate self-reported experiences. A follow-up focus group, involving five volunteers, explored
engagement perceptions in greater depth. Findings revealed that ChatGPT users showed
significantly more acceptance behaviours, while DeepSeek users showed significantly more rejection behaviours. Modification behaviours were slightly more frequent with ChatGPT,but the difference was not statistically significant. Focus group results revealed that ChatGPT was perceived as reliable, supportive, and easy to use, whereas DeepSeek users reported frequent technical issues, low response quality, and frustration. Internal factors such as familiarity with the tool and technical reliability played a more significant role than environmental distractions. Students expressed greater satisfaction and willingness to reuse ChatGPT due to its clarity, responsiveness, and usefulness during revision. The study concluded with practical implications for the integration of AI tools in L2 writing instruction.
