Ask ChatGPT to write something “in the style of Stephen King” and you will now get a refusal — or at minimum a redirect. OpenAI has quietly updated how its flagship chatbot handles style-mimicry requests, blocking prompts that explicitly name a living author as the stylistic target. The change, Ars Technica reports, marks one of the clearest boundaries OpenAI has drawn yet between what the model can technically do and what it will now agree to do. For a hundreds millions users, this is not a minor tweak — it reshapes one of the most popular use cases in creative writing.
The new behavior does not prevent ChatGPT from evoking a style indirectly. Users can still ask for prose that is “terse and atmospheric” or “maximalist and Southern Gothic” without triggering a block. What the model now refuses is the direct name-to-output pipeline: “Write this scene as Cormac McCarthy would.” That specific framing, where a living or recently living writer’s identity is the explicit parameter, now hits a wall.
Why OpenAI Is Moving Now — and What the Legal Pressure Looks Like
The timing is not accidental. OpenAI is navigating a thicket of active and threatened litigation from authors who argue that training on their work without consent — and then producing output that sounds like them — amounts to commercial appropriation. Courts in the United States have not yet settled the question of whether style itself is protectable under copyright law, but the reputational and legal risk of generating text that a living author could credibly claim sounds like their own voice is substantial. Blocking the explicit pathway is a low-cost way to reduce that surface area considerably.
The shift also reflects a broader pattern across the industry. Competitors including Anthropic, whose targets enterprise pipelines, have faced similar scrutiny over how their models handle creative intellectual property. None of the major frontier labs have been entirely transparent about the guardrails they apply to stylistic imitation, but OpenAI’s move is among the most explicit policy signals any of them has issued on the subject so far.
The “Similar Feeling” Loophole and What It Means for Writers
The practical gap between blocking a name and blocking a style is significant, and Ars Technica’s reporting makes clear it is very much intentional. OpenAI appears to be threading a needle: honoring the spirit of author objections by removing the named-attribution pathway, while keeping the underlying capability intact for users who describe a style in terms of technique, tone, or structure rather than identity. A user who knows what makes McCarthy’s prose distinctive — the lack of punctuation, the moral weight in plain declarative sentences, the landscape as moral character — can still invoke all of that without naming him.
For professional authors watching this space, the distinction will feel thin. The concern was never primarily about users typing a famous name; it was about the model’s capacity to absorb and reproduce the texture of someone’s life’s work. That capacity remains. What has changed is the interface layer around it. Whether that satisfies the authors who have publicly objected to AI style cloning — or the lawyers representing them — is a very different question from whether it satisfies OpenAI’s product and legal teams. For now, the two sides appear to be answering different versions of the same question, and the gap between them shows no sign of closing quickly.
