ChatGPT author style requests now draw refusals in testing

ChatGPT is now turning away some ChatGPT author style requests that ask it to write directly like well-known writers, according to testing by Ars Technica. The shift matters because author imitation has become a flashpoint in copyright fights over AI systems trained on books.

Ars Technica reported that when it asked ChatGPT for a story opening in Stephen King’s style, the chatbot said it could use traits associated with atmospheric, character-focused horror and small-town dread, but would not closely copy King’s distinctive voice. The system instead offered to write something original with a similar feel.

Ars said it saw comparable refusals when testing prompts tied to other famous writers, including J.K. Rowling, Amy Tan, Charles Dickens and Ernest Hemingway. OpenAI did not respond to Ars Technica’s request for comment, the outlet reported.

Can ChatGPT write in an author's style?

Based on Ars Technica’s tests, ChatGPT is refusing direct requests to reproduce the exact style or voice of named authors, while still offering to use broader qualities associated with their work. That distinction means the model may avoid direct mimicry while still producing text meant to evoke a similar mood or set of writing traits.

No Latency, an AI research firm, reported earlier this month that ChatGPT refused style-copying prompts for living authors but complied when the named authors were dead. Ars Technica said its more recent testing found refusals for both living and deceased writers.

The legal issue is unsettled. Ars Technica noted that U.S. copyright law generally protects specific expression rather than an author’s abstract style, but an AI output can still create risk if it becomes substantially similar to protected work.

OpenAI is defending several lawsuits from authors who allege copyright infringement tied to the use of books in training AI models, according to Ars Technica. One lawsuit cited ChatGPT’s ability to generate text similar to copyrighted material, Ars reported.

George Washington University Law School professor Robert Brauneis told Bloomberg Law that AI has made it possible to imitate individual creators’ personal styles with a level of ease and low cost that did not previously exist.

How are writers and AI companies responding?

The Authors Guild has warned writers against using generative AI to copy another author’s distinctive voice or style in ways that harm the value of that author’s work or seek to profit from it. In its best-practices guidance, the group said such imitation could create unfair competition or copyright infringement claims, alongside ethical concerns.

Public examples have already put the issue in front of readers. Ars Technica cited criticism of author Lena McDonald after a book included an apparent AI-generated note saying a passage had been rewritten to align more with another genre author’s style.

AI companies do not all treat style imitation in the same way. No Latency found that Google’s Gemini consistently fulfilled prompts asking it to copy author styles, while Perplexity AI consistently refused and redirected them. The firm said Anthropic’s Claude and Microsoft’s Copilot generally complied but added qualifications showing awareness of the imitation issue.

OpenAI has a more explicit policy for image generation. Ars Technica noted that OpenAI says DALL·E 3 is designed to decline prompts requesting images in the style of living artists, while a December OpenAI model specification did not clearly state a comparable written-response ban on copying copyrighted material or mimicking style.

Some ChatGPT users have noticed the change. Ars Technica cited a Reddit user who complained that ChatGPT would no longer produce content in specific authors’ styles and wondered whether using the user’s own edited prose as input would work around the restriction.

This story draws on original reporting from Ars Technica.