You could use AI to help write your book, but you shouldn’t.
Why? The short answer is this: Because your book will then be an amalgamation of other people’s ideas, it will be the most common and obvious of those ideas, and these ideas will be organized and expressed in the most generic manner possible—making the book a 50,000-word cliche. In food terms, a chicken nugget, not a roasted chicken.
The shorter answer is this: Because reliance on AI makes it harder to copyright a book, so publishers will be hesitant to touch it. And even if you self-publish, retailers such as Amazon might purge it. KAA, in fact, will not let its writers use AI and tells its clients not to use AI either.
The shortest answer is this: Because the book wouldn’t be good, and it wouldn’t be yours.
Why you’re tempted to use AI is obvious. Writing a book is a huge undertaking. It takes focus and discipline for draft after draft. It requires craft, thoughtfulness, and endless, exhausting decision-making. It’s a long, lonely process. Authors ask themselves every day, “Will this thing ever be done?” It’s an uncertain process. “Will people like my book?” And it’s an anxious process. “What will this book say about me?” You might tell yourself, I could save a lot of time and pain by using AI—just as you could save yourself a lot of time in the New York City Marathon by taking a subway to the finish line (like Rosie Ruiz did in 1979).
At the moment, the creators of AI have swamped the media with bold proclamations of how revolutionary their product is. The financial press gushes over how much money is being spent on it. Social media is full of images and videos created by AI at the snap of a finger. Many CEOs are foisting it on their employees as the secret to efficiency and the answer to all their problems. Google is using it to diminish search results, which could undermine the integrity of the entire internet. And many people are turning to chatbots for not just the answers to their questions, but companionship and therapy.
In this environment, you’d have to be crazy not to use AI to write your book, right?
To maintain ownership over the book, you might even tell yourself you’ll only use AI for things that supplement the writing process, such as:
- Brainstorming
- Organizing messy ideas
- Building outlines
- Summarizing research
- Tightening sentences
Used with deliberation and intent, AI could make these parts of the writing process faster and less intimidating. But convenience and psychological comfort do not add up to quality.
A good book is not just information dumped into neatly arranged paragraphs. A good book has a voice, a perspective, nuance, and rhythm. These are the result of a particular author’s intentions and choices, the manifestations of an actual human intelligence behind a book.
Those are the very things AI struggles with most because it’s not intelligent. It’s not like Skynet or WOPR. It’s an algorithm, a large language model (LLM) that determines not the best word you should use (the most personal, the most powerful, the most punchy, the most surprising) but the most statistically likely words in response to a prompt. An LLM’s results might seem polished, but so does plastic.
Readers will notice, and they will be put off. They’re coming to the book for what you think, not what some calculator spits out.
The Promise and Pitfalls of Using AI to Write a Book
Using an LLM is appealing for obvious reasons. It’s fast and never gets tired, like we humans. It doesn’t stare dramatically out the window, claiming to be “thinking through the chapter.” It can generate pages of material in minutes. For anyone overwhelmed by the size of a book project, that kind of speed feels like a lifeline.
And sometimes, it is.
If you’re stuck, LLM-generated content can give you a jumping-off point. If you have ten tangled ideas and no structure, it can help you sort them. If you need a rough list of chapter topics, some example prompts, or help clarifying your thoughts, it can be super useful.
The problem is that its strengths are also what make it lousy.
LLM-generated content can appear fluent. That’s not the same as insightful. LLM-generated content can appear confident. That’s not the same as being an expert. LLM-generated content can sound right. That’s not the same as being knowledgeable and providing correct information.
You can get clean, readable text at alarming speed, but that smoothness can fool you into thinking the writing is better than it is. It sounds fine until you look closely. Then you realize it says a lot without actually saying much at all.
But here’s the thing: If you’re using AI to write for you, you probably won’t look at its results closely. You will miss the metaphors that are ridiculous. The facts that are wrong. The arguments that don’t add up. The ideas that are obvious.
You just want the thing done. But in prioritizing your time over the quality of the book, you aren’t respecting your reader.
The fault, as Shakespeare would put it, is not in the stars, but in yourself.
What AI Can and Can’t Do When You’re Writing a Book
Before you decide how much to use AI writing, it helps to be realistic about what it’s good at and where it struggles.
AI as a brainstorming tool for authors
There are two ways to brainstorm. One, use the first thing you think of—often if it’s a bolt from the blue, a lightbulb realization. Two, use the sixth thing you think of because the first five will be obvious and overused.
An LLM could never come up with the former because it is literally designed to not generate anything new. For example, a great idea often comes from melding two unlike ideas. An LLM is incapable of doing that on its own because no one has done it before. Of course, you could prompt an LLM with two unlike ideas to see how it would bring them together, but it could only give you the most generic outputs based on how the ideas are alike, not understanding that it’s the unalikenesses being coordinated that generates greatness.
Don’t cede your originality to something unoriginal.
AI as a drafting assistant for books
Yes, AI can draft, we all know that by now. You can get rough paragraphs, a generic scene, chapter summaries, or a placeholder piece of text. That can be helpful if you need momentum.
But the output usually sounds like exactly what it is: artificially generated language, not the considered, deliberate language that makes readers enjoy and recommend a book.
The generated output may be technically competent and even sound polished. But it lacks texture, surprise, specificity, and the kind of emotional or intellectual depth that makes readers care and connect with human writers.
AI as an editing assistant
If you rely on Word or Google Docs to check your spelling, punctuation, and grammar, you might think such proofreading and line editing is probably where an LLM could be genuinely useful.
Now go back up a page and read that word “unalikenesses,” which should be “unlikenesses,” but which Google Docs didn’t flag as wrong. Just as you can’t trust a word processor to be perfect, you can’t trust an LLM to make your writing better at the sentence-level.
Remember: LLMs learn from vast sets of writing, at least 90% of which, to use Sturgeon’s Rule, is crap. Do you want your novel’s turns of phrase to be rewritten by an LLM so it’s more like crap? Do you want your voice smoothed out so it sounds more like crap? Do you want your sentences tightened so they read more like crap?
No? Then you shouldn’t rely on an LLM to help you.
Why AI is a weak interviewer for books
One of the biggest weaknesses of an LLM in book creation is that it just can’t interview the way a skilled human can.
A great human collaborator knows how to ask follow-up questions and how to hear what you’re almost saying so they can gently dig deeper. Real people know when a memory has emotional weight or when a throwaway comment is actually the heart of the chapter. In other words, they know how to draw out your best material.
An LLM can’t do that because it is not human. As Kyle Reese said, “Listen, and understand! …It can’t be bargained with. It can’t be reasoned with. It doesn’t feel pity or remorse or fear!” So it can’t relate to anyone. Instead, it’s more like the chatbots that are replacing helplines, able to deal with only a limited set of responses.
It can ask pre-programmed or prompted questions, sure. But it can’t listen deliberately like a human does or pick up on:
- Hesitation
- Subtext
- Body language
- Contradiction
- Emotion
- The significance of a small detail that might unlock the whole narrative.
This skill is so important, especially in memoirs, nonfiction, business books, etc. And it’s one of the reasons teams like KAA are so valuable. You get the efficiency people want from modern tools, but paired with human expertise, strong interviewing, strategic guidance, and an actual understanding of how to turn ideas into an amazing, publishable book.
Why LLMs Appeal to People Who Want to Write a Book
Here are some pitfall ideas you should avoid.
An LLM can start the writing process
Can’t get started? Those promoting LLMs say they can generate something for you to react to: a placeholder intro, structural options, a few directions to consider. They’ll be deeply average, of course, but they aren’t meant to be used, just to inspire you to create something brilliant yourself.
In other words, they want you to prompt an LLM to create prompts for yourself. That’s like someone saying, “Want to make $1? Pay me $5, and I’ll give you $1.” Instead, just prompt yourself using the prompts you would have given an LLM.
And the best way out of writer’s block isn’t to get an LLM to write something for you. It’s to simply start writing anything. Once you’re in the flow of writing, direct yourself toward the topic at hand.
AI can help generate outlines and organize ideas
Perhaps you need help organizing your spreadsheets full of references, notes written on Post-Its and bar napkins, ideas on index cards tacked to a corkboard, and material from your journal. You could use an LLM to help you coordinate this chaos.
But it won’t help you make sense of this chaos. The connections it makes will be artificial or obvious, if they are actual connections and not hallucinations. And it won’t be able to cut stuff out that’s confusing an issue or add stuff that clarifies an issue, because an LLM can only work with the material you give it.
You have to make the connections, and that will make your book worthwhile.
Think of it this way. Every detective show inevitably displays the main character standing in front of a bulletin board packed with pictures, names, diagrams, and other evidence, all linked together by a string. But does that string, essentially the detective’s LLM, ever lead to an insight? No. It’s always a new clue, a seemingly unrelated observation, or some stray comment that lets them put things together and solve the crime. The string doesn’t help. Don’t bother with it.
AI can improve efficiency, not authorship
America has a fetish for efficiency. For optimization. The simpler and more straightforward something is, the better.
Great books come from their authors overcoming friction. In fact, any real author will tell you, if something comes too easily, then it must be suspect. So embrace the friction and find your own unique way to deal with it. Don’t remove it with an LLM like some rich guy paying a goldfarmer to level up his character because he doesn’t want to actually play the game he wants to play.
What You Give Up When You Have an LLM Write Your Book
An LLM can’t write in your voice
Voice is not just wording. It’s worldview, rhythm, instinct. It’s the way the world has shaped you and how you’ve responded to it.
AI can imitate patterns of your voice, but don’t expect it to replicate the real thing in a way that consistently holds up across an entire manuscript. You’ll spend more time fixing and rewriting than you would have just writing it from the get-go.
LLM-generated content feels generic
No matter how much you’ve “trained your AI,” it’s always going to give you the most statistically likely version of a sentence, a paragraph, or an explanation. That means the output likely lands in a weirdly polished middle zone, a literary Uncanny Valley—where nothing is wrong exactly, but nothing feels memorable either.
At best, it reads like beige wallpaper. At worst, like the way the movie The Polar Express looks: weird and off-putting.
You might think, “So what? It’s fine. It’s done.” Now imagine that paragraph above being your first review on Goodreads or Amazon. Do you think there will be a second review?
Ethical and copyright concerns around AI writing
This area is still evolving, and that’s part of the bigger problem here.
Writers need to think carefully about copyright, authorship, disclosure, and the legal murkiness around tools trained on existing content. Even when something seems convenient, that doesn’t mean it’s totally risk-free.
Most LLMs today were trained on existing content, such as the millions of eBooks published through Google Play, or fan-fiction posted to Wattpad, or literally anything in the public domain.
So, if you use AI to write your fantasy novel and readers point out that certain excerpts are very similar to popular fantasy novels out there…they might be right, and you may find yourself in court.
Just as there were law firms that sought out people who shared MP3s online and sued them for thousands of dollars, there will be law firms that will, ironically, use AI to find instances of plagiarism and sue the offending authors. Do you want to find out the true cost of efficiency?
AI can flatten emotion and dehumanize storytelling
A book is supposed to connect with readers. That’s literally the whole point. Even in business nonfiction, readers are responding to voice, trust, credibility, tone, and emotional intelligence. In fiction, that emotional depth is even more important because you’re asking readers to connect with made-up people and worlds.
Most LLMs can describe emotions, but they can’t feel them. And readers know the difference, even if they can’t always explain why. They just feel it the way your spouse knows when you’re lying.
Privacy matters when you feed your book into AI tools
When you upload your material to an LLM, some will take it for training purposes. That material might be deeply personal or professionally valuable, and now you’ve made it available to anyone else using the LLM.
You might think, I’m going to put that in my book anyway. Which is fine. That’s your choice. But now other people can put it in their books. Is that your choice?
So, Should You Let AI Write Your Book?
If someone came to your house for a dinner party, and you served McDonald’s, would they be satisfied?
If you asked friends over to see the Chagall you bought, and it was a paint-by-numbers version, would they be impressed?
No.
The basic fact is this: an LLM isn’t an aid. It isn’t a tool. It’s not even a crutch.
It’s merely one way to write a profoundly mediocre book.
Written by Stephen S. Power, Executive Editor at KAA
Stephen S. Power is a nearly 35-year publishing veteran. At Kevin Anderson, he builds books from the ground up: working with authors and ghostwriters, strategizing marketable ideas, developing outlines, editing manuscripts, and writing book proposals for placement with agents or publishers. Several of his proposals have sold for six-figure advances, and many of the books he’s worked on have become New York Times, USA Today and PW bestsellers. He doesn’t use an LLM because nearly 35 years of practical experience have made him far better than them.
Categories:
Uncategorized