Study finds 3 student-AI writing profiles, no performance boost

This paper presents an exploratory mixed-methods study of 44 Hong Kong secondary students completing a Curricular Writing Task with AI chatbots. The researchers analyzed screen recordings to identify ten distinct prompting strategies—including questions, searches, and detailed instructions—that students used to interact with the AI. Through clustering these strategies, they discovered three distinct profiles of human-AI rhetorical load responsibility: AI-dominant (52% of students, where the AI carries most of the writing load), Human-dominant (25%, where students retain control), and Collaborative human-AI (14%, where responsibilities are shared). A MANOVA analysis examined whether these profiles correlated with writing performance in content, language, and organization.

The study's most striking finding was that there was no significant multivariate effect of rhetorical load responsibility on any of the three writing performance dimensions. This suggests that the way students distribute writing responsibility with AI—whether letting AI lead or maintaining human control—does not directly predict the quality of the final text. The authors argue that these prompting strategies and load distribution patterns have important implications for EFL writing pedagogy, particularly concerning student engagement, autonomy, and how teachers design tasks that integrate generative AI. The paper challenges educators to reconsider assumptions about collaboration and authorship when students write with AI tools.