One of the best use cases for AI in writing fiction is as an expert personal research assistant. A common writing tip is to “write what you know.” This doesn’t mean to only write about your personal life, but to write about things you personally know about. That could mean doing research about certain topics you haven’t experienced personally until you are familiar enough to convincingly write about that topic. The problem with research is that it can take a lot of time, which limits the amount of actual creative writing you can do. That’s where AI comes in.

Professional authors already hire assistants to outsource their research so they can spend more time writing the manuscript. And some authors even have such assistants ghostwrite portions of their book for them. Most novice writers do not have the financial resources to hire a human for that role. But AI can perform many of the roles a research assistant would—sometimes even better than a human.

For instance, if I wanted to write a story written in the voice of a medieval monk, it would take me months or perhaps years of research, reading dozens of books to learn about that period in order to get a true sense of the monk’s voice, the way he thought and spoke, in order to convincingly write in that same style. For someone writing a nonfiction history book about that period, such research would be necessary. For someone writing a novel, it might be worth it. But what about just a short story, particularly a satirical piece? Probably not. Spending years of research, reading millions of words, is not worth the time expense for a single 1000-word flash fiction parody. So such a story would never be written (unless a medieval historian decided to do it in his spare time).

But a fiction writer can take a story idea and ask an LLM to write it in the voice of a medieval monk, and it will do so instantly—because it has already “read” thousands of books written by people from that time period and people writing about that era. When it comes to medieval monks (or any other voice you try to mimic), AI may not write at an “expert” level, but it is “good enough” if you’re writing a satirical short fiction story.¹

I wrote before how ChatGPT has no voice of its own and instead mimics the writing of others. Therefore, literary writers should not use AI to write for them. But if the story is meant to be in another person’s voice (like a parody of medieval monk), AI might be better at mimicking them than you are. The kind of research required to learn an ancient monk’s voice (or the voice of any other obscure figure from history) is extremely arduous for any one human writer (or multiple research assistants). Even experts who devote their entire lives to studying a single topic only have so many hours in the day. But AI can do that research for you in an instant. Of course AI still hallucinates and makes mistakes, and it will not be as good as the best human expert in any field, but for fiction writing (pastiche and parody in particular), where precise historical accuracy is less important, AI can be a vital tool.

Using AI as a research assistant opens the door for fiction writers to write many more stories (both in type and quantity) than they ever could have before. I’ve found AI to be the best at flash fiction, and I’ve had a flood of creativity coming up with new story ideas to generate with AI—stories that I could not have written without the aid of AI. That is what human artists should aim to do with any medium: take advantage of its unique qualities. Don’t use AI to replace your unique voice as a writer; only use AI to write stories that could only be written with AI.

Flash fiction is less than 1500 words. It doesn’t take too long to write that many words, but certain historically, technically, and/or scientifically specific stories would take enormous amounts of time spent researching to do it the old-fashioned way. Getting a character’s voice right could take months, if not years. But an LLM can do that research for you and mimic the character in seconds. AI is quite good when mimicking specific types of people (such as a unique profession, someone from a certain time period, or a particular historical figure).

Another type of flash fiction I’ve done with AI is a “found document” story, like the found footage genre in movies. I had an idea for a fake classified CIA file, so I gave an LLM all my ideas about this fictional program, then the AI assembled it in the style of an actual CIA document. AI did all the drudgery of researching how to properly format a classified CIA document, while I got to focus on the purely creative parts of the process.²

LLMs are best at mimicking other writers, so I only use AI for stories that are meant to be in the voice of someone else, not my own. Flash fiction is more about the idea and the style, rather than the development of characters, story, and plot—the things that AI would not be as good at in longer pieces. AI is good at writing in the voice of pre-existing characters or templates, but humans are best at creating the voices of unique characters and inventing new templates. The type of stories I use AI for may be closer to fan fiction than true literature, but fan fiction can be fun and entertaining.

Most of the flash fiction stories I create with AI are in the genre of parody—humorously mimicking the writing voice of someone else. AI excels at the mimicry part, but the human must figure out how to make it funny via the prompt (and subsequent editing). AI opens brand new vistas for comedy writing in the form of eerily accurate parodies of hyper-specific niches. No comedy writer in the past would have spent years studying the private journals of medieval monks just for one joke. But AI can do that for you in a snap.

AI excels at parody but struggles with authenticity. That is why, for my longer literary work, it needs to be fully written by me, a human, to elevate above mere entertainment to art. Plus, due to working memory constraints, LLMs struggle with longer works. So for novels, especially about a personal topic (or when writing about things that I actually do know and only I know), I would not use AI because in those pieces it is my individual voice that is most important—because it makes that story unique. Likewise, this blog post was written entirely by me because it is composed of my personal thoughts and opinions, which AI does not know (for now) and therefore cannot generate. But for gimmicky flash fiction stories that aren’t written in my voice, AI is perfect for mimicking others’ voices. AI enables me to write those stories that I never would have been able to otherwise write.

One example is my story “The Phony on the Carousel,” a review of a fictional sequel to J.D. Salinger’s The Catcher in the Rye. I came up with the idea and outlined the plot, then used an LLM to write it in the style of Holden Caulfield. I’ve read The Catcher in the Rye³, but would have needed to re-read it and closely study the text to get Holden’s voice down. That would have taken me at least a month, during which I wouldn’t have been able to do any other writing—all for a 500-word story. Instead, AI was able to instantly mimic Holden’s voice, then I spent a couple hours refining and editing the text. A story that would have taken a month to write only took me a day.

Another example is “The Chainsaw in the Labyrinth,” a longer story about Javier Milei that was intended to be a pastiche of Jorge Luis Borges, two Argentinians. I used AI to help write in the style of Borges and also to research Milei. The LLM gave me a short piece of flash fiction that I wound up rewriting and expanding upon. It served more as a jumpstart to begin a longer story that I mostly wrote myself. About 90 percent of the final story is composed of my own words.

For longer stories that I am writing entirely myself, AI is still useful as a research assistant. For example, while writing a hard science fiction story set on a moon of Mars, I asked AI specific questions related to the worldbuilding of my particular story, and it gave scientifically accurate answers (such as the gravity on that moon and the effects it would have on the characters). Authors have always asked professional scientists questions to aid their writing, but using AI is faster and easier. It’s like having an expert in every single academic field in your pocket, who is also an expert in your particular manuscript. (Though of course, you need to be aware that AI can sometimes make mistakes.)

There’s a misconception that AI-generated writing takes no effort for the human writer, but there is actually an art to prompting. First of all, the idea itself is the most important aspect—at least in flash fiction. Secondly, it takes a detailed prompt to generate a good story, not just a single sentence. I often write a long paragraph or multiple paragraphs, detailing exactly what I want. With “The Phony on the Carousel,” my original prompt was more words than the final story. After AI generates a first draft, I hone it with additional prompts, reiterating further to get exactly what I want. When I have text from the AI that I’m satisfied enough with, I then rewrite and edit further, entirely myself. It is a true collaboration between AI and human. The percentage of AI versus human text in the final draft differs from story to story, but you cannot say it’s fully AI-generated or fully human-written. It is something else—a new art form.

AI is good for certain types of writing, and writers should use it to their advantage when they can—but you should not use AI for everything. The best use case for LLMs when writing fiction is pastiche and parody. As for original literature, LLMs can serve as an expert research assistant, but the human author should still do the actual writing. AI can write a decent pastiche of Holden Caulfield now, but it could not have written Catcher in the Rye before J.D. Salinger. Developing your own voice as a writer is vital. Let AI write in the voice of others, but only you can write in the voice of yourself.

1. I have written such a parody story in the voice of a medieval monk, but it hasn’t been published yet. ↩︎
2. This story is coming soon to Time Zone Weird. ↩︎
3. It’s actually one of the few books I’ve read twice, but I still would have needed to re-read it a third time. ↩︎