Informality features in AI-generated and human-written academic discourse

Abstract

Academic writing is conventionally characterised by formality, yet recent research highlights a growing use of informal features in academic discourse. This study is a comparative corpus-based analysis of informality in 40 human-written and 40 AI-generated argumentative essays, with the latter produced using ChatGPT under standardised prompting conditions. Drawing on a framework of 12 informality markers, the study examines the frequency and rhetorical function of these features across both corpora. The human-written dataset, drawn from the Louvain Corpus of Native English Essays (LOCNESS), exhibited a markedly higher use of informal features, with first-person pronouns, anaphoric reference, and sentence-initial conjunctions being most common. In contrast, the AI-generated dataset contained significantly fewer informal markers, avoiding most conversational elements such as contractions, second-person pronouns, and fragments, but showing a strong tendency toward sentence-initial conjunctions. Chi-square tests confirmed statistically significant differences for nearly all features, indicating that while human writers frequently blur the boundary between spoken and written registers, ChatGPT consistently suppresses informality, producing a more formal style. Qualitative analysis further revealed that students employ informality to engage readers, convey stance, and create conversational flow, whereas ChatGPT maintains detachment and structural control, sometimes at the expense of personal voice. These findings contribute to applied linguistics research by demonstrating how AI-generated texts may both model and distort academic style. The study underscores the pedagogical implications of integrating AI in higher education academic writing instruction (e. g., EAP and composition courses), particularly in guiding students to critically evaluate stylistic conventions.