ChatGPT author style prompts are now running into a new wall: OpenAI’s chatbot is refusing direct requests to write like named authors and offering to produce text with broader genre traits instead. Ars Technica reported that its tests on Wednesday produced refusals when asking for openings in the style of Stephen King, J.K. Rowling, Amy Tan, Charles Dickens, and Ernest Hemingway.
In one test described by Ars Technica, ChatGPT said it could write atmospheric, character-focused horror with small-town dread, while declining to imitate King’s “exact style” or “distinctive voice.” The model then offered an original passage meant to capture a related mood. That is the mechanism of the new dodge: block the named imitation, translate the request into generic traits, and keep writing.
OpenAI did not respond to Ars Technica’s request for comment, so the company has not publicly explained whether this is a new policy, a model update, a safety layer, or a quieter product tweak. The change is visible in behavior, according to the tests, which is the part users actually encounter.
Why is ChatGPT refusing to copy authors now?
The timing puts the shift in the middle of OpenAI’s continuing fights with authors over copyright. Several lawsuits accuse OpenAI of using books to train its models without authorization. One complaint cited ChatGPT’s ability to produce writing resembling copyrighted text, according to Ars Technica.
U.S. copyright law generally protects a specific expression, not an abstract style. A generated passage can still create legal risk if it is substantially similar to a protected work. That is why the line between “write like this author” and “use broad qualities associated with this kind of writing” is more than product theater, even if the output can feel similar to a reader.
George Washington University Law School professor Robert Brauneis told Bloomberg Law that AI has made it unusually cheap and effective to imitate an individual creator’s personal style. The Authors Guild has also warned writers against using generative AI to copy another author’s voice or distinctive traits in ways that harm the value of that author’s work or seek profit from the imitation. The group says that conduct could invite unfair competition or copyright claims.
How other AI chatbots handle style prompts
A research note from No Latency earlier this month found that ChatGPT refused requests to imitate living authors, while still complying with requests involving deceased writers. Ars Technica’s later tests found refusals for both living and dead authors, suggesting the behavior has changed since that analysis or varies by system conditions.
No Latency also reported that rival systems handle the same problem unevenly. According to the firm, Google’s Gemini consistently followed author-style imitation requests. Perplexity’s models refused and redirected those prompts in a way similar to ChatGPT. Anthropic’s Claude and Microsoft’s Copilot complied, while adding caveats that signaled some awareness of the imitation issue, the researchers said.
OpenAI has already drawn a clearer line in image generation. The company says its Dall-E 3 system is designed to reject requests for images in the style of living artists. Ars Technica noted that OpenAI’s model specification published in December did not state an equally clear written-response ban on mimicking style or reproducing copyrighted material.
The change will annoy some users who built writing workflows around named-author prompts. A Reddit user in a writing-with-AI forum complained that ChatGPT now refuses to generate content in the style of specific authors and said earlier, highly specific prompts had produced the desired results. That frustration is the predictable user-side cost of a system that was trained to imitate patterns and is now being told, selectively, to stop doing the most obvious version of that trick.
This story draws on original reporting from Ars Technica.
