At leisure away from office duties, software engineer Petra Ferraz de Novaes finds herself immersed in imagined realms built by artificial intelligence.
Novaes, a 36‑year‑old software engineer based in Brazil, has been passionate about storytelling throughout her career. Yet crafting even the tiniest decisions—whether tone, grammar, or nuanced character development—seemed daunting. Consequently, in 2024 she turned to an AI model to bring one of her concepts to life.
“She enjoys receiving the off‑script twists that arise from her inputs, often leading the plot in surprise directions. When chatting with WIRED, Novaes remarks, “What I love is the unpredictability.” While many describe AI as prone to hallucination, she believes those quirks deliver precisely the effect she seeks.”
Today, Novaes devotes hourly sessions to storycrafting, employing SillyTavern—a platform that executes open‑source language models locally. After sketching a premise—such as an elven warrior confronting the Amazon queen for her throne—the model populates the missing details, turning scaffolding into fully realized scenes.
Within one session, Novaes jumps fluidly among multiple mythic settings, ranging from worlds born by her own imagination to borrowed universes inspired by Star Wars, Pokémon, and Mass Effect. Occasionally she expands a standalone tale into a novel‑length monologue intended for publication, while most of the time she engages in private role‑playing adventures where the quest begins, is completed, and moves on to the next. As she put it, “The limit is simply your imagination”.
“Even though participants are aware the narratives were generated by an AI system, they show little concern,” explains Melanie Walsh, an assistant professor at the University of Washington Information School and co‑author of the research. “Indeed, the novelty likely fuels interest.”
Consequently, the literary world has faced notable controversies concerning AI‑assisted writing. A prominent case involved a publisher withdrawing a multimillion‑dollar contract after suspicions of AI involvement, while another removed a novel from view amid similar accusations. These episodes underscore a lingering belief that machine‑produced fiction undermines the inherent value of human storytelling.
However, scholars contend that reader response reveals a growing appetite for AI‑crafted narratives, shifting focus away from concerns over artistic diminution.
By analyzing over half a million prompts supplied to ChatGPT via the WildChat public dataset, the team discovered that more than one‑third of conversations incorporated fictional content—ranging from prose and poetry to fan‑fiction, erotica, and interactive role‑play.
The overwhelming share of creative requests originated from a minority of avid users who craft and refactor stories with relentless dedication.
One particular individual once tasked ChatGPT with generating countless iterations of an eclectic Doki Doki Literature Club fan‑fiction scenario, repeatedly across several months. Even when adjusting for unusually heavy engagement, the analysis assigns an estimated seven percent of total chatbot participants to regular fictional storytelling.
Nonetheless, the WildChat corpus carries constraints. Its prompt pool remains unpersonalized, preventing demographic profiling. Additionally, contributors voluntarily disclosed their messages, so the set does not necessarily mirror the broader ChatGPT user base. A 2025 collaborative study with the National Bureau of Economic Research reports only 1.4 percent of ChatGPT prompts correspond to fiction tasks—a stark contrast to the researchers’ higher figure, which excludes role‑play contexts and cannot verify erotic themes.
Also Read
- Does estrogen protect against dementia? And if so, who should take it?
- IT Team Laughs as Colleague Holds Power Button to Prevent Outage
- Big AI sets out its terms for regulatory capture and calls it ‘Pace the frontier’
- Insight Partners’ Devin Parekh on Prioritizing Diversification Amid the AI Investment Frenzy
