ChatGPT Stopped Writing in the Style of Rowling, King and Hemingway

For several days now, ChatGPT has been refusing requests like "write a story in the style of Stephen King" or "create a poem like Czesław Miłosz." The change in the OpenAI chatbot's behavior was first noticed by Reddit users, and was independently confirmed by tests from Ars Technica and Engadget journalists. The company has not issued any official statement on the matter.

The blocking mechanism is simple but consistent. Instead of faithfully reproducing a given author's voice, ChatGPT offers text "inspired" by the general features of their work - genre conventions, mood, narrative techniques - while explicitly refusing to reproduce the "exact style" or "distinctive voice."

What the refusal looks like

In tests described by Android Authority, when asked for text in the style of Stephen King, ChatGPT offered a text drawing on recognizable features of his writing - conversational narration, mounting dread, psychological tension - but explicitly stated it would not reproduce the author's "exact style or distinctive voice." A similar scenario played out in Polish tests by Spider's Web, where attempts to get text "like Sapkowski or Miłosz" ended in an identical refusal.

Rather than mimicking King's exact phrasing, this captures the hallmarks of his style: conversational narration, creeping dread, psychological tension - ChatGPT, in a response quoted by Android Authority

Why now

OpenAI has not publicly explained the reasons for the change, but the legal context is clear. The company is defending itself against several class-action lawsuits from authors, including the Authors Guild, George R.R. Martin and John Grisham, who allege it used their books to train language models without consent or compensation. Copyright law in the United States protects the specific content and form of a work's expression, not "style" itself as an abstract concept - but generated text can infringe copyright if it becomes sufficiently similar to a protected original.

Pressure on the industry is also rising due to financial precedents. Anthropic agreed to pay $1.5 billion in a settlement with book authors in a similar case over training data - one of the largest settlements in the history of US copyright law, and a reference point commentators explicitly link to OpenAI's decision.

We've never before had a tool this good and this cheap at copying a creator's personal style - Robert Brauneis, law professor, George Washington University, quoted by antyweb.pl

Competitors take a different approach

According to an industry audit by No Latency, cited by Android Authority and Spider's Web, Google Gemini still carries out requests to imitate specific authors' styles without restriction. Claude and Microsoft Copilot fulfill such requests but add legal disclaimers. Perplexity AI, like ChatGPT now, refuses to imitate the style of both living and deceased writers. That means there is no single industry standard yet - each company draws its own line between inspiration and copying.

A loophole remains

The block applies only to explicitly named authors. As IBTimes UK notes, a user can still get text that sounds like King's or Rowling's prose by describing the desired tone, mood and narrative technique without naming the author - for example asking for "a horror story set in a small American town" instead of "a text in the style of Stephen King." This shows the change is more symbolic and preventive than technically airtight.

For Polish creators and ChatGPT users, the change means that previously popular prompts like "write like Sienkiewicz" or "create a poem in the style of Szymborska" may stop working, even though the work of both authors is subject to different copyright rules in Poland than in the US. OpenAI has not announced separate rules for markets outside the United States, so for now the restriction applies globally and uniformly.

The case fits into a broader dispute over the boundary between inspiration and copyright infringement in the era of generative AI, one now playing out not just in courts but in the models' own settings. Future rulings in cases against OpenAI, along with decisions by other language model providers, will show whether blocking the names of specific authors becomes an industry standard or remains a one-off OpenAI experiment.