ChatGPT now refuses direct requests to imitate authors’ styles

Lead

According to testing by Ars Technica, ChatGPT has begun refusing prompts that ask it to directly mimic the writing style of famous authors. When asked for an opening in the style of Stephen King, for example, the system said it could write with atmospheric, character-driven horror and small-town dread, but could not closely imitate King’s distinctive voice.

That change may sound subtle, but it matters. OpenAI is still facing lawsuits from authors who argue that large language models were trained on copyrighted books and can generate text uncomfortably similar to protected works. The new behavior appears designed to move ChatGPT away from explicit “voice cloning” and toward more generalized creative guidance.

Key points

• Direct imitation is being blocked: Ars found similar refusals for prompts involving both living and deceased authors, including J.K. Rowling, Amy Tan, Charles Dickens, and Ernest Hemingway.
• The model still writes creatively: Instead of copying an author’s exact style, ChatGPT offers to use broad qualities such as tone, atmosphere, pacing, genre conventions, or narrative mood.
• The legal line remains blurry: In the United States, copyright generally protects specific expression rather than an abstract style. But an AI output could still become legally risky if it is substantially similar to a copyrighted work.
• Platforms differ: The article cites a No Latency analysis finding that Gemini tended to comply with author-style imitation requests, Perplexity consistently refused and redirected them, while Claude and Copilot took more qualified middle-ground approaches.
• Users are noticing: Some longtime ChatGPT users have complained that the change disrupts workflows built around highly specific style prompts.

Why it matters

For OpenAI, this looks like a product-level guardrail aimed at reducing copyright and unfair-competition exposure. The company is not necessarily saying that all stylistic influence is unlawful. Rather, it is trying to avoid responses that openly promise to reproduce the identifiable voice of a particular writer.

For authors, publishers, and AI-assisted writers, the message is more practical: using AI to study structure, genre, pacing, or atmosphere may remain acceptable, but using a living or famous author’s voice as a reusable template is becoming a sensitive practice. The Authors Guild has already advised writers not to use generative AI to deliberately mimic the unique styles, voices, or distinctive attributes of fellow authors in ways that could harm the value of their work.

The hard problem is that the boundary is not technically clean. “Broad qualities” and “exact style” are not fixed categories. A model may claim it is only capturing a similar feeling, while readers may still recognize a strong resemblance to a particular author. Over time, platform policies, court decisions, and professional norms will likely shape where the acceptable line for AI-assisted writing lands.

Source: Ars Technica AI