ChatGPT Refuses to Copy Author’s Style
July 28, 2026 – 9:27 pm
(Image by Canva)
Asking ChatGPT to write a chapter in Stephen King’s style now results in a refusal. Instead, it offers to craft something with "the hallmarks of atmospheric, character-driven horror and small-town dread," while maintaining its own distinct voice.
OpenAI has recently changed how its chatbot handles author imitation.
Previously, the model could write in the style of both living and dead authors, but now it declines requests for imitating specific writers due to copyright concerns.
When Did the Change Happen?
Two weeks ago, ChatGPT would still imitate dead authors but refused to copy a living author’s style. By July 27th, however, the chatbot also began declining requests for works in the style of deceased authors like Charles Dickens and Ernest Hemingway.
Why the Shift?
The key shift lies not in the ability to imitate, but in when it refuses. While copyright law doesn’t protect a writer’s style, OpenAI is now drawing the line at any work still under copyright. This change appears to be a strategic move to avoid legal issues given the numerous copyright lawsuits against OpenAI from authors and publishers who allege their works were used to train ChatGPT.
Legal Implications
While US copyright law protects specific expressions of ideas, not styles, ChatGPT‘s subtlety in refusing writer’s "exact style" while offering a general "feeling" could still be significant. Imitation can become infringing if it becomes "substantially similar" to the original work, which legal scholars suggest is a potential risk for AI-generated content.
Variations Across Labs
Unlike other AI models, ChatGPT isn’t alone in its stance. A recent audit by No Latency found varying approaches among different labs:
- ChatGPT and Perplexity refuse and redirect on living author prompts.
- Anthropic’s Claude and Microsoft’s Copilot comply but attach originality caveats.
- Google’s Gemini complies outright, despite facing its own copyright suit.
There seems to be no uniform rule across AI labs, suggesting that these decisions are policy choices rather than technical limitations.