Abstract
Writing is central to how students develop and demonstrate communicative competence, yet that development is increasingly difficult to observe as GenAI enters the writing process. This paper establishes a pre-GenAI stylistic baseline of student academic writing and characterises how unedited GenAI responses to the same assessment prompts differ from it. Using stylometric and writing analytics techniques, we analysed over 118,000 words collected in 2021 from 18 higher education students in a software engineering subject. Study 1 used the seven students who completed all four tasks to establish a within-cohort reference: the size and direction of stylistic change in the same students’ writing across two semesters and two rhetorical genres. Study 2 paired each of 59 authentic submissions with a length-matched, zero-shot response to the identical prompt from three large language models. Student writing moved toward shorter words, fewer rare terms and markedly higher readability. Unedited model outputs moved in the opposite direction on each measure, gaining lexical sophistication through a narrower working vocabulary at substantial cost to readability; the human–GenAI gap in readability exceeded the students’ own shift between semesters.
Keywords
Generative AI, academic writing, stylometry, writing analytics, AI in education.
