Writing with a Reader in Mind

The Rhetorical Gap Between Genuine EFL Student Essays and LLM-Generated Essays

Abstract

Recent research has highlighted the potential of large language models (LLMs) such as ChatGPT to support the development of ESP writing skills. However, concerns have also emerged about learners delegating writing tasks to these tools, potentially undermining learners’ motivation and autonomy. LLMs’ capacity to generate human-like text challenges the ability to ensure the authenticity of EFL students’ writing. Some studies have investigated LLM-generated text characteristics, comparing them with student writing. However, research on LLMs in EFL education remains limited. This paper seeks to address this gap and advance critical understanding of LLM-generated text. It reports on a small-scale study comparing three corpora: genuine undergraduate EFL student essays, suspected LLM-generated essays, and texts produced by ChatGPT 4.0 and DeepSeek-V3. The study examined patterns of interactive and interactional metadiscourse in the essays. Combining quantitative analysis with qualitative interpretation, it was found that although interactive features have limited explanatory power in distinguishing between LLM-generated text and genuine student essays, LLMs place strong emphasis on explicit cohesion, coherence and logical organization, favoring an objective and factual style. Most importantly, while they approximate human stance-taking, LLMs show very limited capacity to model audience and engage readers. This pattern was mirrored in the essays suspected to be LLM-generated. In contrast, genuine student writing was found to display a more personal tone and substantially greater reader engagement, regardless of target language proficiency.