ChatGPT declines to imitate named authors' writing styles

Ask ChatGPT to write a paragraph in Stephen King's voice, and you now get a polite decline. As Ars Technica reported, OpenAI's chatbot has started refusing direct requests to imitate named authors, replying that it can conjure "the hallmarks of atmospheric, character-driven horror and small-town dread" but "can't write in Stephen King's exact style or closely imitate his distinctive voice." The reporting says the same refusal extends to J.K. Rowling and Amy Tan, and, in Ars's own testing, to deceased writers like Charles Dickens and Ernest Hemingway, though a separate analysis from No Latency found the model still complied for some deceased authors.

The timing matters. OpenAI is in the middle of a stack of author copyright suits, and one of them, per the coverage, specifically cites ChatGPT's "uncanny ability to generate text similar to that found in copyrighted textual materials." A product that says on the record it will not clone a named author's voice is a different exhibit in a courtroom than one that eagerly produces a passable King pastiche on request. U.S. copyright doctrine mostly protects specific expression rather than intangible style, but as the reporting notes, outputs that end up "substantially similar" to a protected work can still infringe, so the refusal is less about erasing capability than about controlling what the product will admit to doing.

The awkward part, and the one the reporting flags, is that a refusal at the top-level chat is not the same as an inability. If a user pastes in a few paragraphs of a target author and asks for "more like this," or leans on a fine-tuned or open-weights model, style-copying stays trivial. Legal scholar Robert Brauneis, quoted in the coverage, puts the underlying problem plainly: "We've never had a situation in which this personal style of individual creators could be imitated as well and as inexpensively as we now have with AI." The refusal message is a policy layer over a capability that has not gone anywhere.

What the reporting does not give you is a firm rollout date, whether the change applies uniformly across ChatGPT's tiers and the API, or which specific suit prompted it. Take the specifics as reported, not settled. The direction, though, is clear enough: expect other frontier providers to add similar refusals on named creators, expect plaintiffs' lawyers to test whether the guardrail is real or cosmetic, and expect the next round of the fight to be about whether "style" itself starts to attract a form of protection U.S. copyright has never really given it.