LLMs struggle with author impersonation, favoring text reuse over style mimicry

A new study published on arXiv evaluates the ability of six major large language models, including Gemini 3 Pro, Claude Opus, and GPT 5.2, to impersonate individual authors. Researchers found that while Gemini 3 Pro demonstrated the strongest mimicry, it achieved this primarily through extensive sentence and phrase reuse, rather than genuine stylistic imitation. The study suggests that training set consistency is a more significant factor in impersonation success than author distinctiveness, and that LLMs often exhibit model-specific preferences that separate their generated text from target author styles. AI

IMPACT Highlights the limitations of current LLMs in genuine stylistic imitation, with implications for detecting AI-generated text and forensic analysis.

RANK_REASON Research paper published on arXiv evaluating LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Claude Opus
• DeepSeek-3.2
• Gemini 3 Pro
• GPT 5.2
• Grok 4.1 Fast
• Shunichi Ishihara

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →