ChatGPT Now Refuses to Mimic Famous Authors' Styles
16 Aug 2026
OpenAI's ChatGPT has quietly changed how it handles one of its most popular creative-writing tricks: asking the model to write "in the style of" a famous author. The chatbot now declines direct requests to mimic the styles of both living and deceased writers, including J.K. Rowling, Amy Tan, Charles Dickens, and Ernest Hemingway.
What changed
An analysis by No Latency published earlier this month found that ChatGPT complied with style-copying requests for deceased authors. Current testing shows the opposite: ChatGPT now refuses such requests for both living and deceased authors alike. Sources do not clarify exactly what technical or policy change drove this shift, or when precisely it occurred — only that the behavior has flipped within a short window.
Why this is happening
The timing lines up with OpenAI's ongoing legal exposure. The company is fighting multiple lawsuits brought by book authors alleging large-scale copyright infringement by models trained on their work. One lawsuit specifically cites ChatGPT's "uncanny ability to generate text similar to that found in copyrighted textual materials."
Professor Robert Brauneis noted that personal creative styles have never before been imitable as well and as cheaply as with current AI — a capability that sits at the center of the legal dispute over whether style mimicry constitutes infringement.
Open questions
Several details remain unclear from current reporting:
- What specific technical or policy change triggered the refusal behavior
- Whether the block extends to indirect or paraphrased style requests, not just direct ones
- How many authors or lawsuits are involved, and their current legal status
- Whether this applies across all OpenAI products or just ChatGPT
- The methodology and sample size behind the refusal testing
- Whether competing models like Claude or Gemini show similar behavior
Sources differ on the underlying facts here in one notable way: the earlier No Latency analysis found compliance with style requests for deceased authors, while current reporting shows refusal for both living and deceased authors. This appears to reflect a genuine behavior change over time rather than a factual contradiction — but no source explains the mechanism behind it.
Why founders should care
This shift is likely a signal, not an isolated tweak. For founders building generative AI products, a few probabilistic takeaways stand out:
- Content policy risk is probably rising. If OpenAI is tightening style-mimicry guardrails in direct response to legal pressure, other providers may follow, and further restrictions seem plausible.
- API reliability for style-dependent features may be at risk. Startups whose products depend on stylistic imitation — ghostwriting tools, personalized content generators, fan-fiction platforms — could face sudden behavior changes with little warning, as this case shows.
- Legal precedent could reshape the space. The lawsuits against OpenAI may set outcomes that affect how any AI tool handles stylistic imitation going forward, potentially raising compliance costs or restricting features industry-wide.
- Inconsistent enforcement raises reliability questions. A model complying with a request one month and refusing it the next suggests safety and content policies at major AI labs may still be in flux, which founders should factor into product roadmaps.
The opportunity side
Not all of this is downside. The tightening around direct mimicry could open room for:
- Writing tools that offer stylistic inspiration rather than direct imitation, sidestepping legal exposure while still serving similar user needs
- Compliance-focused AI content tools built explicitly around respecting creator style rights
- New tools that help authors detect or protect their stylistic signatures from AI mimicry — a defensive product category that didn't clearly exist before
For now, the practical takeaway is that founders building on top of ChatGPT or similar models should have contingency plans if style-mimicry capabilities are core to their product, since this case demonstrates such features can be restricted with little public explanation.