Abstract The rapid rise of ChatGPT has generated considerable anxiety among teachers concerned that students might turn to large language models to write their assignments. This AI-powered language model is able to create grammatically accurate and coherent texts, thus potentially enabling cheating and undermining literacy and critical thinking skills. While research has begun to examine linguistic and rhetorical differences between AI-generated and human-authored texts, far less is known about how large language models construct authorial stance, a central component of argumentative writing and a key indicator of academic literacy. This study compares stance markers in argumentative essays written by British undergraduates with those generated by ChatGPT on the same topics. Using a corpus-based approach grounded in an established stance framework, we identify systematic differences across epistemic, attitudinal, and self-referential resources. Findings show the AI texts contained significantly fewer stance expressions, deployed a narrower lexical repertoire, and projected markedly weaker authorial presence than the student essays. We attribute these differences to the heterogeneous language data used to train ChatGPT and its underlying statistical algorithms. The results highlight both the limitations of AI-generated academic prose and the pedagogical potential of using such texts to develop students’ critical AI literacy and rhetorical awareness.
Jiang et al. (Wed,) studied this question.