Dissecting AI discourse

Abstract

This study investigates the discourse characteristics of AI-generated argumentative essays designed to simulate student writing, comparing them with human-authored texts to explore linguistic variation and its implications for academic discourse. While large language models (LLMs) such as ChatGPT demonstrate impressive fluency and coherence, questions remain about their capacity to engage in nuanced argumentation and the critical reasoning expected in academic contexts. Employing Biber’s (1988) multidimensional (MD) analysis, this study examines systematic differences between LLM-generated and human-written essays across dimensions of register variation, informational density, syntactic complexity, and persuasion. The results reveal that ChatGPT essays are markedly more informationally dense, context-independent, and abstract, yet exhibit reduced interpersonal engagement and overt persuasion compared with student essays. The paper concludes by reflecting on broader pedagogical and ethical implications for the integration of AI tools into academic writing and corpus-based research.