How to Write Your Book with AI Without Losing Your Soul

The hardest thing for me to write about these days is AI.

Not because it’s constantly changing, though it is. 

Not because there are a thousand “hot takes,” and I wonder how I can contribute meaningfully to the conversation. But that’s not why it’s hard to write about. 

It’s hard to write about AI because I feel like I’m being curmudgeonly, and I don’t like feeling curmudgeonly. I am neither old enough nor bad-tempered enough to have earned that label. 

But whenever I give people honest advice about writing with AI, I run the risk of crushing their dreams a little.

Here’s what usually happens. I mention that I use AI myself, and with my clients, and people light up. They lean in. They want to know the hack, the cheat code, the prompt that unlocks the thing they haven’t been able to get AI to do, no matter how many times they’ve tried.

Some of these people would even call themselves superusers. And yet, when it comes to writing their book, AI hasn’t given them what they wanted.

I don’t have a cheat code for them. And even if I did, I couldn’t just hand it over.

The cheat code is expertise, not a prompt

The thing that makes AI work — for me, and for the handful of clients I’ve watched use it well — isn’t a clever prompt. It’s the critical thinking underneath it. For me, that’s a decade of writing and editing more than 30 books, plus the transferable expertise I’ve gained from earning a PhD in philosophy and teaching students how to think.

That clarity of thought is what makes the tool useful. Without it, AI is just a very confident collaborator with no idea whether it’s helping or hurting you.

So no, there isn’t a magical way to use AI that you’re missing. It’s not about finding the right prompt. It’s about being intentional, knowing what you actually like, and using AI only where it makes sense for you. 

I’m not down on AI. I use it. I have a lot of fun with it, actually. But I want people to stop expecting it to be a miracle worker, because it isn’t one. I think, slowly, people are starting to come around to that.

What I learned editing an AI-assisted manuscript

This spring, I edited the first full book a client had written largely with AI. It was, without exaggeration, one of the hardest editing jobs I’ve ever taken on. The writing wasn’t bad, exactly, but it wasn’t hers. Here’s what that experience confirmed for me.

Writing expertise still matters. If you don’t already know what makes something sound “off,” AI can’t tell you either. For example, my client asked it to make her book funnier. It did, technically. But it wasn’t funny in the way she’s funny. It was funny in the way a stranger imitating her might be. AI doesn’t know your timing, your specific brand of self-deprecation, the joke you’d actually make in that moment. Only you know that, which means only you can catch when it’s missing.

You still have to validate your idea. If you don’t know what makes something resonate with your reader — what you want them to do after they read it, what they will find valuable and worth their time — AI isn’t going to figure that out for you. It will happily generate confident, well-structured prose around an idea that has no real substance. That’s arguably worse than an idea with obvious gaps because it’s harder to see the problem underneath all that polish.

Nothing replaces the discipline of the struggle. If you don’t wrestle with your idea somewhere along the way, you end up with slop — competent-sounding, forgettable slop. I’ve tried asking AI to push back on my thinking, and honestly, I haven’t had much luck with it. It’s built to be agreeable, and agreeable is not the same as useful. What actually works is testing your ideas somewhere with real friction: on social media, in a workshop, in front of people who will tell you when you’ve lost them.

More words in less time is not automatically a good thing. AI let my client produce a full manuscript faster than she ever could have on her own. But that also meant the book arrived twice as long as it needed to be, and I spent so much time rewriting that there wasn’t room left for the deeper editing the book actually needed. Speed is great, but clarity is golden. And clarity almost always means fewer words, not more.

AI is not the problem

I want to be clear. My client didn’t hand her outline over to AI and let it generate the whole book unsupervised. She was involved every step of the way. She had ideas, opinions, a sense of what she wanted the book to say. By most definitions, this was AI-assisted writing, not AI-generated.

And it still didn’t work the way she hoped.

That’s the part I think gets missed in most conversations about “using AI the right way.” The issue was never really whether she was assisting the AI or letting it generate things wholesale. The issue was that she didn’t have the expertise to assist it well. She didn’t know how to catch the moment her humor stopped sounding like hers (or she overlooked it because of AI blindness or because she was in a hurry). She couldn’t see, in the moment, that more words weren’t the same as a better book.

That’s not a failure of effort or involvement. It’s the same skill gap I mentioned earlier — the particular type of critical thinking that comes from years of writing and editing. Being “hands-on” with AI doesn’t automatically mean you’re directing it well. You can be fully in the driver’s seat and still not know where you’re going.

What this means if you’re writing your own book

If you’re an author trying to use AI well, here’s my advice:

Use AI where it’s genuinely useful — as a mirror that reflects your thinking back to you, an avatar of your reader, a tool for moving faster through the parts that don’t require your particular judgment or where you’ve already thought long and hard. Don’t use it to skip the parts that feel hard.

Know your own voice well enough to recognize when something doesn’t sound like you because AI is still not great at pointing that out for you.

Validate your idea before you scale your output. Speed only helps once you know what’s actually worth writing.

Find your own version of the struggle — run a workshop, talk it through with someone who could become a trusted reader, talk about your idea during a podcast interview — because that friction is doing something AI can’t replicate. It’s making the idea, and the book, actually yours.

AI is a tool. A genuinely good one, when you know what you’re doing with it. It doesn’t perform miracles. The soul of the book has to come from you.

Image by magnific.com on Freepik.