I’ve published more than a hundred books and ghostwritten over fifty for other people, and I use AI most days I work. That sentence gets me in trouble in both directions. Half the writing world thinks admitting it makes me a fraud. The other half thinks it means I’m about to tell you the machine can write your book. Neither is right, and the space between those two reactions is where the answer lives.
The answer starts with a better picture of what the thing is. AI isn’t a writer. It’s a digital assistant, and the moment you treat it like an assistant you hired, everything about how to use it well falls into place.
How do you treat AI like a writing assistant?
Think about how you’d work with a person you brought on to help you write. You’d learn fast where they’re strong and where they’re weak. Maybe they’re a sharp researcher who pulls together background in an afternoon that would take you two days. Maybe they’re good at rough structure and hopeless at a finished sentence. You wouldn’t fire them for the things they cannot do. You’d stop handing them those things.
That’s exactly the relationship. You direct the assistant. You say, go research the history of this topic and give me the major turning points. You say, I’m stuck on how to open this chapter, show me three ways in so I can react to them. You say, I have a mess of notes, help me build an outline I can argue with. You say, my brain has quit for the day and I cannot find the next paragraph, give me a bad draft I can fix.
Every one of those is a real assistant doing real assistant work. None of them is the assistant writing your book.
You’re still the brain. You still do the writing. The judgment stays with you. What belongs, what’s true, what sounds like you, what earns its place on the page. That’s the part the assistant cannot do and never will. This is the same line I draw in whether AI can write your book, and the answer there’s the same as the answer here. It can help you write your book. It cannot write it.
Where does AI break?
The reservations aren’t small print. They’re the reason you keep the assistant on a short leash, and every one of them has cost a writer somewhere real money and real embarrassment.
It hallucinates. It’ll hand you a fact that’s wrong and a citation that doesn’t exist, delivered with the same calm confidence it uses for the things it gets right. I’ve watched it invent a source, a study, and a quote, and never once flag that it was making them up. It doesn’t know it’s making them up. If you put its facts into your book without checking every one, you’re signing your name to whatever it invented.
I have watched it invent a source, a study, and a quote, and never once flag that it was making them up.Share on X
I go deeper into this in why bad inputs break AI projects, but the short version is that the machine is a confident liar and you’re the fact checker.
It drifts. Give it a long enough job and it forgets what you told it at the start. It wanders off your instructions a little at a time, until what it’s producing no longer matches what you asked for. I wrote a whole piece on AI drift because it’s the failure people notice last and pay for most.
It flattens your voice. Hand it your writing and ask it to help, and it’ll quietly sand you down toward the average of everything it’s ever read. Your odd rhythms, your specific word choices, the small things that make a sentence sound like you and not like anyone else, it planes them off without asking. What comes back is smoother and deader.
And it makes everything the same. This is the one that surprises people. Ask AI to help with a nonfiction book and every chapter opens the same way, closes the same way, runs the same length, and sits at the same flat middle level of quality. Ask it to help with fiction and you get the same thing, every chapter the same size and the same wishy-washy shape. The spine of the book comes out uniform, and uniform is another word for mediocre.
A real book has chapters that breathe differently because the material demands it. The machine doesn’t know the material demands anything.
Why does AI struggle with hard or dark writing?
There’s a whole category of writing the assistant simply won’t do well, and it’s the category that matters most in a lot of books. Anything confrontational. Anything dark. It cannot write a sex scene worth reading, and it won’t write most violent scenes at all.
You can push it, and it’ll grudgingly produce something, the way you can push a nervous employee into a task they hate. What comes back is watered down and wrong. You spend more time fixing it than you would have spent writing it clean.
This is where the whack-a-mole starts. If you lean on AI for the material it’s bad at, you end up with a manuscript that reads wrong in ways you cannot immediately name. A scene that should land hard lands soft. A moment of real conflict comes out weirdly polite. You keep poking at it, fixing one dead spot only to find another, never quite understanding why the manuscript feels off.
The answer is that you asked the assistant to do the work it was never going to be able to do, and the fix isn’t more prompting. The fix is to write those parts yourself.
Is it ethical to use AI to write?
The AI models learned from other people’s work. David Baldacci sat in front of a Senate subcommittee in July 2025 and described watching his son ask ChatGPT to write a plot in the style of a David Baldacci novel. In about five seconds it produced three pages carrying plot lines, character names, and narrative pulled from across his books.
He said it felt like someone had backed a truck up to his imagination and stolen everything he’d ever created. That’s a real objection from a serious writer.
The tool exists whether or not any individual writer approves of it. Say you’re writing a kids’ book and you ask AI for an outline, and it hands you something that reads a lot like an R.L. Stine Goosebumps book. You could also walk into a library, pull a Stine off the shelf, and follow that outline yourself.
Writers have studied and borrowed structure from other writers for as long as there have been books. The machine did it faster and from a bigger pile. That’s a real problem and it’s baked into the tool.
So if you choose to write with AI, you take the caveats that come with it. One of them is that somebody may notice, and somebody may write you a scathing review for it. Fair enough. But somebody was going to write you a scathing review anyway. No matter what you write or how you write it, someone out there will hate it and say so at length. That’s the job.
It’s not a reason to avoid a useful tool, and it’s not a reason to pretend the tool is cleaner than it is.
Are writing tools like ProWritingAid and Grammarly any better?
The dedicated writing apps are a different category from the big chat models, and one of them earns a spot in my toolkit. ProWritingAid is a good tool. It does a solid job proofreading and a solid job on spelling, and I think it’s better than Grammarly at both. Its overused-word finder is useful, because once you can see your own crutch words you can fix them going forward.
I use it and I recommend it, with one caveat. The AI-suggestion side is a work in progress, the same as the AI portion of everything else right now. Lean on the proofreading and spelling. Treat the rewrite suggestions the way you treat any AI output, as something to check instead of accept. Do that and the tool does right by you.
Hemingway is an even lighter option, and it does its narrow job well because it stays in its lane as a proofreader and readability checker that doesn’t try to rewrite you. Between a tool like ProWritingAid for the careful proofread and Hemingway for a quick readability pass, you have what you need for the mechanics. What you shouldn’t do is hand any of them a loose rein over your finished prose.
When you use AI for a grammar pass, tell it plainly: don’t change a word except the ones I point at. Give it free rein on your manuscript and it’ll go off rewriting the manuscript while your back is turned, flattening you the way it always does.
How should a writer use AI?
Point it at the assistant work and keep it off the writing. Let it research. Let it rough out an outline you’ll tear apart. Let it break a block by handing you something bad enough to argue with. Let it check your grammar under strict orders. Let it help you find phrases for a title or back-cover copy, which is marketing, not the book. And keep it far away from your voice, your facts before you’ve checked them, your hard scenes, and any decision about what the book is.
The writers who get burned are the ones who forgot they hired an assistant and started treating it like the author. The writers who do well are the ones who stayed the brain, kept the machine on the tasks it’s good at, and did the writing themselves. If you want the fuller map of where the tool helps and where it hurts, I laid it out in what AI is good at and the six places it’ll break your work.
And if you’d rather hand the manuscript to a human who uses these tools the careful way and takes responsibility for every word, that’s the ghostwriting work I do. Either way, the rule is the same. It’s a tool. Use it wisely, and it earns its keep. Use it as a shortcut around the work, and it’ll show.