ChatGPT has started refusing requests to write in the style of well-known authors whose works remain under copyright, according to reporting by Ars Technica published July 27, 2026, and independently confirmed by Engadget the following day.

When Engadget asked ChatGPT to copy Agatha Christie's style, the chatbot declined, stating: "Agatha Christie's works are still under copyright, so I can't provide text that closely imitates her distinctive style." ChatGPT then offered a middle path, saying: "I can certainly write something with those characteristics, while remaining original." Ars Technica's article, titled "ChatGPT stops cloning famous writers' voices but may capture a 'similar feeling'," framed this as a behavioral shift where the model declines direct stylistic imitation but still attempts to produce text sharing broadly similar qualities.

A system prompt is the set of instructions a chatbot receives behind the scenes that shapes how it responds to users. A fine-tuning adjustment, by contrast, involves retraining parts of the model on new data to change its behavior at a deeper level. Engadget attributed the original discovery to Ars Technica using a straightforward present-tense attribution ("Ars Technica reports") and then ran its own tests, finding results consistent with the earlier reporting. Firstpost, an Indian news outlet, also covered the development on July 28 under the headline "OpenAI's ChatGPT no longer mimics famous authors' writing styles."

A study by No Latency, cited in Ars Technica's original reporting, found that Google's Gemini consistently complied with requests to copy an author's style, in contrast to ChatGPT's refusals. The comparison establishes that this refusal behavior is not uniform across what the industry calls frontier models — the most capable AI systems available from major developers. At least one competing system still honors such requests without a copyright-based objection.

The distinction ChatGPT draws is narrow but consequential. The model will not produce text that "closely imitates" a copyrighted author's "distinctive style," but it will write something with the general characteristics associated with that author. Where the line falls between capturing a "similar feeling" and producing a derivative imitation is not something the chatbot articulated in the exchange Engadget documented. It is also not clear from the available reporting whether OpenAI deployed an explicit policy change, a system-prompt update, a fine-tuning adjustment, or some combination of what are called guardrail mechanisms — the automated checks that prevent a model from producing certain kinds of output. The reporting describes what users observe, not the underlying implementation.

What is verifiable is the copyright reasoning the model itself surfaces. ChatGPT's refusal is not generic safety boilerplate; it explicitly invokes copyright status as the basis for declining. That is a meaningful signal about how OpenAI's current guardrails reason about stylistic imitation, even if the model's stated rationale does not necessarily reflect the precise technical mechanism producing the behavior.

The broader context here is a model voluntarily declining a category of request that has been routine for large language models since their consumer deployment, and doing so on copyright grounds specifically. Authors and rights holders have filed suits against AI developers over training-data usage and output similarity, and a model that pre-emptively refuses to imitate a living or recently deceased author's style changes the practical calculus for users who relied on that capability. At the same time, the model's offered alternative — writing "something with those characteristics, while remaining original" — preserves a degree of stylistic approximation that may or may not satisfy either users or rights holders depending on how closely the output tracks the source author's voice.

The competitive divergence with Gemini adds another dimension. If one major model refuses a class of request and a competitor does not, users for whom author-style imitation is a feature rather than a bug have a clear reason to migrate. Whether that creates competitive pressure in either direction, toward stricter refusals or toward continued compliance, will depend on factors the current reporting does not address, including user churn data and any legal or regulatory developments.

For developers building applications on top of these models, the shift is a reminder that guardrail behavior can change between model versions or updates, and that capabilities available through one provider's API — the programming interface developers use to connect their own software to a model — may not be available through another's. The No Latency study's finding that Gemini still complies where ChatGPT does not means that cross-provider parity on creative tasks cannot be assumed.

The Ars Technica article and Engadget's independent confirmation establish that the behavior is reproducible as of late July 2026. Whether it persists, tightens further, or relaxes is an open question that will depend on OpenAI's own decisions and the broader legal landscape around AI-generated content and copyright.