STATISTICAL PATTERNS AND STYLISTIC OVERLAP:

Abstract

As artificial intelligence becomes increasingly involved in the production of academic, creative, and literary writing, the distinction between human-authored and AI-generated text has become more difficult to establish. In response, AI-detection tools have been widely adopted, particularly within educational institutions, where their results often carry significant academic consequences. However, growing evidence suggests that these systems frequently produce inconsistent outcomes and rely on statistical markers whose reliability remains questionable. This study investigates three such markers commonly associated with AI detection namely repetition, Shannon entropy, and variation in sentence length to examine whether they can meaningfully distinguish AI-generated writing from human literary text. The proposed study compares two texts: a human-written abstract of George Orwell's Animal Farm and an AI-generated text produced from a prompt based on the same narrative. This comparison allows for an examination of whether statistical markers commonly associated with AI writing can reliably distinguish between the two. Using Python-based computational analysis, the study measures patterns of repetition, lexical predictability, and sentence-length variation while interpreting the findings through literary and stylistic perspectives. Consequently, the study argues that statistical markers cannot be treated as independent or definitive evidence of AI authorship. By bringing literary stylistics into conversation with computational approaches, the study aims to challenge the assumption that AI-generated and human-written language exist as clearly separable categories. Instead, it proposes to demonstrate that significant stylistic overlap exists between the two, raising important questions about authorship and the reliability of some common AI-detection systems.