DOES AI EDITING RESHAPE ACADEMIC STYLE? A LEXICAL DENSITY ANALYSIS OF AI-EDITED VERSUS ORIGINAL MEDICAL ARTICLES

Abstract

Artificial intelligence has witnessed a notable expansion recently, for being incorporated into a multitude of academic and professional aspects. This study investigates the lexical density of AI-edited medical abstracts compared with their original versions. Using a quantitative corpus-based comparative design, 30 PubMed-indexed medical article abstracts published between 2019 and 2021 were selected and edited using a standardized prompt in ChatGPT-5. Lexical density was measured with AntConc following Ure's (1971) framework. The findings revealed highly similar lexical density levels between the two datasets. Human-written abstracts obtained a mean lexical density score of 62.06 (SD = 3.83), whereas AI-edited abstracts recorded a mean score of 62.10 (SD = 3.12). Although the AI-edited versions displayed slightly greater consistency in lexical distribution, the difference in lexical density was statistically negligible. The results indicate that AI-assisted editing has only a limited effect on lexical density and preserves the informational load and overall lexical characteristics of medical article abstracts.