OpenAI quietly changed ChatGPT so it declines prompts asking it to write "in the style of" a named author, offering to capture the general feeling instead — a move read as a hedge against its copyright lawsuits.
2026-08-09 · by Moe Ameen
In late July 2026, reporters and users noticed that ChatGPT had quietly started refusing prompts that ask it to write "in the style of" a specific named author. Asked to imitate a particular writer's distinctive voice, the model now declines the direct request and offers a workaround instead — writing something with the same atmosphere, tone, and techniques while keeping the text "original." For a Stephen King prompt it responded that it could "write with 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." For Agatha Christie it cited copyright directly: her works "are still under copyright, so I can't provide text that closely imitates her distinctive style."
The refusal was first observed on living writers — names like J.K. Rowling and Stephen King — while pastiches of long-dead authors such as Charles Dickens and Ernest Hemingway still went through. Within days that gap closed: testers found ChatGPT dodging requests for deceased authors too, citing "distinctive style" rather than copyright status. There was no formal announcement. OpenAI did not publish a changelog or a statement explaining when the behavior was introduced, and its December model spec carried no equivalent written rule for text — which is why the change reads as a quiet retrofit rather than a stated policy.
The context is legal. OpenAI is defending itself against multiple copyright suits from authors and publishers over how its models were trained, and "write in the style of [author]" is precisely the kind of output that plaintiffs point to. Blocking the direct ask is a low-cost way to reduce the most quotable examples of alleged mimicry, even though the underlying capability is unchanged.
Because the capability is unchanged, a simple loophole remains: describe the qualities you want instead of naming the author. A prompt for "a horror story set in a small American town, ordinary characters, a growing sense of dread" produces a King-adjacent result without ever asking for King. Competitors handle the same requests differently — Google's Gemini has complied with named-author style prompts directly, while Claude and Copilot have tended to comply with disclaimers attached. Treat the exact scope as a moving target: OpenAI can adjust these refusals at any time without notice.
If your content depended on asking ChatGPT to sound like a famous author, this change is a nudge toward something better: your own voice, defined once. [Kompozy](/) doesn't borrow a named writer's style — it runs on a **Persona Brief**, a saved profile of how *you* write (diction, cadence, phrases to use, words to ban). Every output the engine generates is governed by that brief, so there's no "in the style of" prompt to refuse in the first place. You're not imitating Stephen King; you're compounding a recognizable voice that's yours to keep, across every post.
That voice then rides the whole pipeline. From one idea Kompozy produces [Text Posts](/glossary/output-buckets), [Persona Tweets](/glossary/persona-tweet), carousels, a [Persona Short](/glossary/persona-shorts) narrated by your avatar, a blog article, and a newsletter — all held to the same Persona Brief and the same look via [HyperFrames](/glossary/hyperframes) templates — then schedules them across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot). Under the hood it draws on Claude, OpenAI, and Gemini, so you're never exposed to a single vendor quietly changing what it will and won't write. The lesson of this policy shift is simple: don't build on a rented style that a lab can revoke overnight — build on a voice you own and distribute everywhere.
It stopped honoring the direct request. As of late July 2026, ChatGPT declines prompts that ask it to write "in the style of" a specific named author, saying it can't closely imitate that writer's distinctive voice. It still offers to write something with the same tone, atmosphere, and techniques while keeping the text original — so the underlying capability is intact.
OpenAI didn't publicly announce it, but it comes as the company defends multiple copyright lawsuits from authors and publishers. "Write in the style of [author]" is exactly the kind of output cited in those disputes, so refusing the direct ask reduces the most quotable examples of alleged mimicry. Its December model spec had no equivalent written rule for text, which is why it reads as a quiet retrofit.
Yes. Because only the direct "in the style of" phrasing is refused, describing the qualities you want still works — asking for a small-town horror story with ordinary characters and mounting dread produces a Stephen King–adjacent result without naming him. Other assistants also differ: Gemini has complied with named-author prompts directly, while Claude and Copilot tend to add disclaimers.
Build on a voice you own rather than one you borrow. In Kompozy, a Persona Brief defines how you write once — diction, cadence, banned words — and governs every generated post, so there's no "imitate a famous author" prompt for a model to refuse. That same voice carries through text, carousels, avatar video, blogs, and newsletters, scheduled across the eight social platforms plus blog and email.