“`html
When Petra Ferraz de Novaes is not at work, she is likely engrossed in crafting a fantasy narrative with the assistance of artificial intelligence. The 36-year-old Brazilian software engineer has always had a passion for storytelling. However, the intricate decisions involved in narrative creation—from style and syntax to character development—often felt overwhelming. In 2024, she decided to enlist the help of an AI model to bring one of her ideas to life.
Novaes was captivated by the eccentric and sometimes unpredictable interpretations of her prompts, which led the narrative down unexpected paths. “What fascinates me is the unpredictability. They say AI hallucinations are quite common. But in this case, it’s probably just what you want,” Novaes remarked in a recent conversation.
Nowadays, Novaes spends countless hours generating stories with AI, utilising a programme called SillyTavern to run open-source models on her computer. She begins with a concept—such as an elf warrior battling the Amazon queen for her throne—and observes as the model expands on the tale.
In a single session, Novaes navigates between multiple fantasy realms, some of her own creation and others inspired by franchises she loves, including Star Wars, Pokémon, and Mass Effect. Occasionally, she crafts standalone stories that resemble short novels, sharing them online. At other times, she engages in private role-playing games. “It can be endless. You find a quest, complete it, and move on to the next. The only limit is your imagination,” she explains.
The Community of AI Fiction Enthusiasts
Recent research indicates that many people share Novaes’s passion. “They know the stories are generated by an AI model, and they don’t mind. In fact, that seems to be a significant part of the allure,” mentions Melanie Walsh, an assistant professor at the University of Washington’s Information School and co-author of the study.
Recently, the publishing industry has faced a series of controversies regarding the alleged use of AI by professional authors. A publisher cancelled a $2 million book deal; another withdrew a novel from shelves; readers claimed that an award-winning short story was AI-generated. These incidents reflect a prevailing sentiment that readers are averse to machine-generated narratives, believing that using AI to create fiction diminishes the art form.
However, Walsh and her colleagues argue that this uproar has overshadowed reader behaviours and what they may signify regarding interest in AI-generated fiction.
After analysing over 500,000 prompts from ChatGPT, extracted from a public dataset known as WildChat, researchers discovered that more than a third of interactions with the chatbot involved some form of fictional generation, including prose, poetry, fan fiction, erotic literature, and role-playing scenarios.
The researchers found that the vast majority of fiction requests came from a small group of “advanced users” who obsessively generate and refine stories. In one instance, a single user prompted ChatGPT to create variations of the same fan fiction narrative, set in the world of the anime game Doki Doki Literature Club!, thousands of times over several months. Even accounting for this intensive usage, researchers estimate that around 7% of chatbot users engage in fiction generation.
Understanding the Appeal of AI-Generated Literature
The WildChat dataset does have its limitations. Due to the anonymity of requests, it is impossible to gather demographic data. Furthermore, participants had to consent to their requests being made public, which means the group may not accurately reflect the general population; OpenAI has stated that “WildChat is not a representative sample of all ChatGPT users.” A 2025 study conducted by OpenAI and the National Bureau of Economic Research suggested that only 1.4% of ChatGPT requests pertain to fiction—a figure considerably lower than the researchers’ estimate. OpenAI’s figure does not account for role-playing, and the company could not confirm whether it includes erotic literature.
Despite these discrepancies, Walsh and her colleagues maintain that it is evident that a dedicated sub-group of readers is generating personalised and self-directed fiction.
Researchers believe that some individuals are drawn to the ability to rapidly and repeatedly develop a story, while others appreciate the chance to steer the narrative without being constrained by an author’s choices. “Many qualities of AI models differ from those of human writers, which appeal to at least a segment of users. These models do not make judgments; they can produce stories instantly; they are nearly inexhaustible,” Walsh explains.
For enthusiasts of erotic literature, AI provides an opportunity to fully immerse themselves in the action and interact with characters they create. “It’s like a digital choose-your-own-adventure book, but with infinite possibilities,” describes Razrien, a 40-year-old man from Indiana, who requested anonymity to candidly discuss his sexual fetishes.
Razrien uses AI to create fantasy characters that dominate him in role-playing scenarios: a “large, angry rat wife” and an “eight-foot-tall, feisty wolf mom,” among others. He has learned which models best cater to specific fantasies; Google’s Gemma 4 is currently his favourite for all aspects. “I’m attracted to big, angry women who slap me around. Some models are perfect for that. Occasionally, I dip a toe into the furry universe. Some models do great for that,” Razrien explains.
Concerns Among Experts Regarding Reader Exposure
The ability to generate fiction raises concerns that it may limit individuals’ exposure to writing styles that could challenge their comfort zones. However, AI-generated fiction may also challenge “some older, idealistic notions of literature, which posit that there is a vital communication between the writer and the reader,” argues Neel Gupta, a doctoral student at the University of Washington and co-author of the study.
Gupta adds that for many readers, the relationship is much more straightforward: “It’s about obtaining entertainment, getting what they want.”
Recent AI models struggle to produce elaborate prose, complex characters, and stylistic embellishments, performing better in genres reliant on recurring narrative arcs and devices. Moreover, they often fail to maintain coherence across entire books. Nevertheless, researchers suggest that as models improve, some of these shortcomings may be addressed.
A possible outcome could be a tiered publishing system, where a vast array of AI-generated books and self-published fiction caters to the average reader, while a smaller segment of human-written fiction satisfies niche interests in experimental forms or narrative ambiguity. “Students and professors of English Literature, as well as some individuals in the publishing world, have always had a sort of niche interest,” Walsh notes.
For Novaes, creating stories with AI has been more of an introduction to reading than a substitute for conventional novels. In addition to generating her own tales, she now seeks out AI-generated creations from others. “I’ve never been much of a reader. But now I read the stories of others. They are the only novels I skim through.”
“`