I Am Begging You Not to Use AI to Write Your Book

Look, I am not here to lecture you about the ethics or environmental impact of AI.

Nor am I, as a longtime publishing professional, going to rehash any industry-specific arguments about copyright, data scraping, or dishonest use of LLMs.

There are plenty of other sources you can refer to in order to read more in depth about any of that.

No, I am here to literally beg you, O Self-Publishing Indie Author, not to use AI to write or edit your book manuscript.

Please don’t do it.

Let me tell you why.

But First!

Let me also say that I don’t actually hate AI as a technology.

(Ducking as apples are being thrown at me by those who are vehemently against it.)

I’m an Aquarius, OK? I love technology and am always super excited to learn about cutting-edge programs and systems. And I feel like there are some genuinely useful things that AI, even LLMs, can do for us.

A PERSONAL AI ANECDOTE

For example, a year or two ago, my husband inherited a precious family heirloom, his great-grandmother’s Lithuanian prayer book that had been smuggled out of her home country in the early 20th century.

My husband teaches college-level English and has written a monograph and has coedited an essay collection and has authored many academic articles. So his first impulse when he got his hands on the prayer book was a researcher’s impulse: to find out more about its history, about where and why and how these texts were produced and what they meant to the communities they were made for.

He began reaching out to historical societies in Vilnius, hoping to get in touch with someone who could give him more information. The few people who responded were very kind and shared what they could. But short of turning this into a formal research project, there wasn’t much else he could do to satisfy his curiosity.

He certainly didn’t have any kind of budget to hire a Lithuanian translator to go through the book, page by page, with him.

So what did he do? He put a bunch of the text into ChatGPT for a rough translation, so he could at least have an approximate sense of what he was looking at.

Is this something that would pass muster among hardcore genealogists? Probably not. But he never aspired to be one and has never planned to do anything formal with this as a project. So this was an easy and convenient way for him to get some additional clarity on a deeply meaningful piece of his family’s history.

OTHER INTERESTING USES FOR AI

I think there are also some potentially interesting uses for AI out there for folks who have already produced a ton of their own writing and are prepared to train an agent exclusively on their own body of work so that the resulting text is directly informed by their own style and vocabulary. (I learned a lot about this from Natalie MacNeil.)

Or, there’s something like this “Personal Encyclopedia” project that gives you the tools to create a Wiki for significant events in your own personal life without needing to turn your info over to shady data-mining companies. 

I’m not even sure that I object to measured, intentional use of AI tools to help with a broad overview evaluation of one’s own manuscript while still moving big chunks of it around and getting clear on the structure. (Erin Servais of AI for Editors has a bunch of trainings on how exactly to do that, tools which look particularly helpful for novelists who need to generate character lists, plot timelines, and so forth.) 

And of course it would be naive of me to assume that we’re ever going to be able to fully disentangle ourselves from the places in our digital lives where AI is being foisted upon us in tools that we’re already using. I know I certainly haven’t done all the things necessary to turn AI off within my various Google accounts and inside the Microsoft programs like Word that I use literally every day of my work life as a professional book editor. (The Gotham Ghostwriters’ “AI Guidelines for Ghostwriting” are super helpful in this regard.) 

No, what I’m really saying (after all this very human throat-clearing) is that it will break my heart if you explicitly use AI to write, re-write, or “edit” your book manuscript. 

And I’m not saying this from some theoretical headspace based on assumptions and projection. I’m saying this because I’ve worked on a handful of projects recently that used LLMs at some point in the writing process before they arrived at my desk. And honestly, they were a huge drag to work on.

Pain Point #1: Hallucinated Research

We’ve all heard plenty of stories about AI’s tendency to hallucinate, to just straight-up invent facts and data and then give a mild “whoops, my mistake” response when confronted by the human user about it. Well, somehow this very common knowledge is still not stopping people from taking ChatGPT’s results at face value and presenting them as original “research.”

In a recent book project I was hired to copyedit, I spent a lot of time fact-checking the author’s sources, mostly to confirm that the citations were formatted properly. I was also reviewing the content of the quotes that had been used in the text to make sure that the wording was accurate too. I’d made it well over halfway through the manuscript and had encountered no major problems until I got to a tranche of quotes clustered across a few chapters that I was suddenly having a devil of a time confirming.

I slowly realized, with dawning horror, that they must have been hallucinations pulled from an LLM search. The quotes sounded legit; they seemed like the kinds of things the people being referred to would have said. They certainly bolstered the argument that the author was making. Alas, they were pure bullshit.

I politely restrained myself from telling the author that they were clearly bullshit.

Instead, I couched my queries along the lines of, “Hmm, I’m having trouble verifying these quotes in my own online research; would you be able to share the exact URLs with me?” When the author couldn’t, in fact, hand them over, we ended up cutting the quotes altogether (whew) and replacing them with some similar but actually verifiable sources.

The author ended up paying me for a couple extra weeks of work because of this, though!

I mean, I was of course grateful that they recognized the over-and-above effort that went into my fixing these issues for them. But also . . . if they’d just taken a few extra steps during their own composition process to confirm the quotes before they ever plugged them into the manuscript (or, y’know, had not used AI for research at all), they could have saved some money. (And saved me a bit of a headache.)

Pain Point #2: word slop

Another self-publishing author I worked with not too long ago was very forthright about disclosing his AI use in the front matter of the manuscript itself. In his prefatory material, he described using it as a thinking partner that helped him organize his ideas and expand parts of his arguments that he’d maybe initially given short shrift to.

Fair enough. I was willing to concede the possible usefulness if he was convinced that his LLM of choice had helped him get his message out into the world in a way that he felt more comfortable with.

But as I continued to read his manuscript, I started to have this strange sort of echoing sensation in my head. Why did the writing somehow seem so familiar? I finally realized it sounded familiar because the LLM-assisted writing “style” was effectively identical to the writing in the other project with the hallucinated source quotes! Oh my god, of course.

When a writer is just working with an LLM that’s trained broadly on all general, miscellaneous writing (and not the author’s own previous original material in a more honed framework), then it’s going to default to the now extremely recognizable (and honestly cringe-induing) AI “voice.” Take a look at Ameerah Bad'r’s article “Indicators of AI” or engineer Ossama Chaib’s invaluable repository tropes.fyi, and you’ll see examples of exactly what I mean. 

This manuscript had ’em all. The negative parallelism. The “Not X. Not Y. Just Z.” sentence structure. The egregious use of colons and em dashes (often within the same sentence, ugh). And on and on.

I am absolutely not kidding when I say that I developed a horrible sore throat after my first full-length editing session trying to untangle some of these lexical and syntactical nightmares. I can only chalk it up to my throat chakra being absolutely ravaged by the harsh energy of that algorithmically generated text.

The tragedy for me here wasn’t about the extra work that I had to do to re-edit the mess. It was that the author had felt like these tropes were somehow helping him with his voice. When, as far as I was concerned, I couldn’t hear his voice at all. I could only hear the machine.

Why I Love Working with Self-Publishers

After two decades spent in the traditional publishing world, I first started helping self-publishing authors as a side hustle, just because I loved their enthusiasm. I loved their urgency and their conviction, their big stories and bold messages. It was a thrill to be working with folks who maybe didn’t consider themselves writers first and foremost yet didn’t let that stop them from sharing their wisdom, their expertise, their life experience. 

I love the honesty of a perhaps idiosyncratically constructed sentence if it’s clear that it’s coming straight from someone’s heart. I always tell prospective new clients that they should never be embarrassed or self-conscious about not writing with perfect grammar or spelling. Cleaning up grammatical infelicities or imposing stylistic consistency across a full manuscript is literally what they’re hiring me to do as a copyeditor. It is my job to know that stuff so that an author doesn’t have to. 

My aim is never to make the writing so perfect that no squiggles show up anywhere in a grammar-checking program. (A grammar-checking program, which, ahem, is likely being run by AI these days anyway.) My job is to make the author’s voice come through as clearly as possible; my copyediting clean-up efforts are truly meant to be invisible.

So what’s going to start happening if more and more authors start pouring their manuscripts, at whatever stage of completion, into ChatGPT or Claude or Gemini?

The writing is all going to start to sound the same.

That same dull ba-dum, ba-dum, ba-dum repetition. The same pithy sense of false urgency. The same word choices even. 

So as you’re embarking on your self-publishing passion project, ask yourself:

Are you writing to make yourself indistinguishable from every other person coughing up content into an LLM? Or are you writing to share some deep, human part of yourself in an effort to connect with the deep, human part of other readers? 

As far as I can tell, originality of voice is only going to become more valued and valuable as time goes on and LLM usage is increasingly normalized.

And again, I don’t say this from a place of absolute judgment that not using AI is somehow the most correct moral stance.

It’s just that I can guarantee that your writing is going to be more interesting without the intrusion of AI style. Your editor will thank you for skipping the slop, sure, but more importantly, so will your intended audience.

Because then they’ll be able to really hear you sing.