A Study on Linguistic Clichés in Reflective Essays by Highly AI-Dependent College Writers

Abstract

This study empirically investigates the stylistic and structural characteristics of reflective essays written by students highly dependent on artificial intelligence (AI)—a genre meant to capture unique individual experiences and subjective insights. Using a self-reported dataset in which students voluntarily disclosed their degree of AI reliance upon submission, thirty essays with an AI dependency rate exceeding 30% were selected for analysis. While existing literature has predominantly favored quantitative research on English-language data, this study qualitatively examines how AI intervention transforms thought structures and narratives in Korean subjective texts. The analysis reveals that, due to the next-token prediction and local optimization mechanisms inherent in Large Language Models (LLMs), such texts display a distinct “conceptual emptiness” in which surface-level fluency coexists with inner superficiality. Specifically, an abnormal excess of cliché-ridden metaphors and a standardization of sensory and synesthetic expressions led to data homogenization, while a lack of macro -level lexical coordination resulted in the overexposure of deictics and discourse markers such as *gyeolguk* and *ije*. Syntactic frameworks-negative constructions and the “Not A but B” contrastive structure-were mechanically repeated, and narratives relied on conclusion-oriented storytelling paired with formulaic, hopeful closures. These findings offer concrete linguistic criteria enabling educators to perceive patterns of AI intervention in student assignments, serving as a foundational resource for ensuring academic integrity and establishing practical guidelines for AI use in the generative AI era.