LLM Authorship Impersonation Study Shows Gemini 3 Pro Reliance on Copying
November 5, 2026
Evaluation of six LLMs including GPT 5.2 and Gemini 3 Pro reveals that effective authorship mimicry remains difficult. While Gemini 3 Pro showed the highest mimicry scores, analysis indicates it achieved this primarily by verbatim sentence reuse from training data.
HOW THIS AFFECTS YOU
●
researcherYou should investigate the distinction between true stylistic mimicry and training data memorization in evaluation frameworks.