What question did this study set out to answer?

The aim is to investigate how well large language models can mimic the writing styles of notable authors.

April 7, 2026

Decoding AI authorship: can LLMs truly mimic human style across literature and politics?

Key Points

The aim is to investigate how well large language models can mimic the writing styles of notable authors.
Employs a zero-shot prompting framework for generating synthetic text.
Evaluates output using BERT and XGBoost for text classification.
Integrates LIWC markers, perplexity, and readability indices for analysis.
AI mimicry is identifiable, with XGBoost achieving accuracy similar to neural classifiers.
Perplexity emerges as the key metric showing differences between AI and human text.
LLMs can approximate some syntactic features but lack affective depth and stylistic variety.

Abstract

Abstract Amidst the rising capabilities of generative AI to mimic specific human styles, this study investigates the ability of state-of-the-art large language models (LLMs), including GPT-4o, Gemini 1.5 Pro, and Claude Sonnet 3.5, to emulate the authorial signatures of prominent literary and political figures: Walt Whitman, William Wordsworth, Donald Trump, and Barack Obama. Utilizing a zero-shot prompting framework with strict thematic alignment, we generated synthetic corpora evaluated through a complementary framework combining transformer-based classification (BERT) and feature-based machine learning (XGBoost). Our methodology integrates Linguistic Inquiry and Word Count (LIWC) markers, perplexity, and readability indices to assess divergence between AI-generated and human-authored text. Results demonstrate that AI-generated mimicry remains highly detectable, with XGBoost models trained on a restricted set of eight stylometric features achieving accuracy comparable to high-dimensional neural classifiers. Post-hoc feature importance analysis indicates that perplexity is the most influential discriminative metric, suggesting systematic differences in the distributional regularity of AI outputs relative to the greater variability observed in human writing. While LLMs exhibit distributional convergence with human authors on low-dimensional heuristic features, such as syntactic complexity and readability, they do not yet fully replicate the nuanced affective density and stylistic variance inherent in the human-authored corpus. By isolating measurable statistical divergences in current generative mimicry, this study provides a structured benchmark for LLM stylistic behavior and offers insights for authorship attribution in digital humanities and social media contexts.

Mark Helpful

Bookmark

Relay