Step-by-step guides on prompting, styles, and getting the most out of AI image generation.
Read the guides
ChatGPT has begun refusing direct requests to copy a named author's writing style, OpenAI confirmed — but the model will still generate text that captures a writer's "broad qualities," a distinction that matters for AI-assisted storytelling and may carry legal weight.
The new behavior is specific: ask ChatGPT to "write in the style of Cormac McCarthy" and it refuses. Describe McCarthy's stylistic fingerprints — sparse punctuation, bleak landscape prose, fragmented interiority — and the model complies, producing text that reads recognizably like his work without ever invoking his name. That gap is the crux of the policy.
According to Ars Technica, OpenAI framed the change around capturing a writer's "broad qualities" rather than cloning their voice — language that echoes copyright doctrine while stopping short of resolving it. U.S. copyright law has never protected style, only specific expression, so the legal line between "cloning" and "capturing broad qualities" remains genuinely unsettled.
For creators using ChatGPT to draft narrative content — character backstories, dialogue, lore documents for AI-generated worlds — the practical impact is modest but real. Prompts that name authors now bounce. Prompts that describe style attributes still work. Anyone who has built a workflow around named-author style references will need to translate those references into descriptive language instead: swap "write like Ursula K. Le Guin" for "write in a measured, anthropological third person with understated moral weight."
That extra step is also, arguably, better prompting. Describing the actual qualities you want produces more consistent results than invoking a name, because the model's internal representation of any given author is noisy and varies by genre and era. The new restriction may inadvertently push users toward more precise, transferable prompt craft.
"New behavior capturing a writer's 'broad qualities' could have legal implications."
— Ars Technica
The legal exposure is the more consequential angle. OpenAI's wording suggests the company is trying to thread a needle: block the surface-level act of impersonation while preserving the functional utility that makes ChatGPT useful for writers. But if a plaintiff can demonstrate that "broad qualities" output is substantially similar to protected expression, the distinction may not hold in court. Anthropic's recent $1.5 billion author settlement — which paid roughly $3,000 per book but left the underlying training-data question unresolved — signals that these disputes are far from over.
This policy change is specific to ChatGPT's text output. Image-generation models, including those available through platforms like Charmloop's AI image generator, operate under separate guidelines and are not currently subject to equivalent named-artist style blocks — though pressure in that direction has been building across the industry. The distinction matters: a creator building a visual style inspired by a specific illustrator can still work iteratively with image models in ways that text models are now restricting.
For AI-art creators whose practice involves narrative as well as image work — writing character lore, scene descriptions, or worldbuilding text alongside generated visuals — the workflow shift is worth noting now. Descriptive style prompting, rather than named-author prompting, is becoming the more durable approach across both modalities. The guides section at Charmloop covers prompt-craft techniques that translate directly to this kind of descriptive, attribute-based style targeting.
OpenAI has not published a detailed policy document outlining exactly which names trigger the block or how the "broad qualities" threshold is defined — meaning creators will likely discover the edges through trial and error as the rollout continues.
