What’s the Problem If a Good Book Is Written With AI?

Someone asked this in a writing group last week, and meant it sincerely. It usually gets asked rhetorically — as a way of ending the argument rather than having it. This time it wasn’t, so it deserves a real answer.

Here’s mine: the question contains a hidden assumption, which is that good is the only variable that matters. It isn’t. There are at least four separate questions tangled up in that sentence, and only one of them is about quality.

Is the book any good. Do you own it. Do you have to say how it was made. And did you get what you wanted out of writing it.

Those come apart. A book can be good and unownable. It can be good, owned, and still something you’re required to disclose. Sorting them out is more useful than arguing about whether AI is good or bad, which is an argument nobody wins because the two sides are usually answering different questions.

Start with the one that isn’t a matter of opinion

If an AI generated the actual expression in your book, you may not own it. That’s not a taste position. It’s where US law currently sits.

In March 2026 the Supreme Court declined to hear Thaler v. Perlmutter, leaving standing the DC Circuit’s holding that the Copyright Act requires a work to be authored in the first instance by a human being. The human authorship requirement stands. Works generated by an AI system without meaningful human authorial contribution aren’t eligible for copyright registration.

The Copyright Office has been more specific about where the line falls, and this is the part most writers haven’t heard. Writing a prompt is not authorship. The Office’s position is that choosing among AI outputs doesn’t get you there either — in their words, the selection of a single output “is not itself a creative act.” What’s protected is the perceptible human expression in the finished work. Your selection, coordination and arrangement of material can qualify. The generated material itself doesn’t.

So the practical answer to “what’s the problem if the book is good” is: you might have written a good book that you cannot register, cannot enforce, and cannot cleanly sell the rights to. That problem doesn’t care how good the prose is. A publisher’s contract will ask you to warrant that the work is yours, and you want that to be true.

I’m an editor, not a lawyer, and the Office is explicit that these are fact-specific, case-by-case determinations. If real money is involved, ask someone qualified. But the shape of it is clear enough to plan around.

Amazon already drew the line, and it’s not where most people assume

If you self-publish, this stopped being theoretical a while ago. Amazon’s KDP guidelines split AI use into two categories, and the split is instructive.

AI-generated means text, images or translations created by an AI tool. You must tell Amazon when you publish. Note the part that catches people out: this applies even if you substantially edited the output afterward. Editing does not launder the origin.

AI-assisted means you made the thing and used AI to refine, error-check or improve your own work, or to brainstorm ideas you then developed yourself. No disclosure required.

Read those definitions again, because the line isn’t drawn at how much AI you used. It’s drawn at what the AI produced. Did it produce the words, or did it help you with words you produced? That single question decides which category you’re in.

Janis is right, and more right than she may realise

Someone in that thread answered the question in one line: AI is a good tool, but the author has to be the source and editor.

That’s the whole thing. It’s also, near enough, the legal test — arrived at from craft instinct rather than case law, which is the best evidence I know that it’s the right line.

Source means the expression originates with you. Editor means you’re the one exercising judgment over what stays, what goes, and what order it goes in. Copyright protects the perceptible human expression and the selection and arrangement of material. Amazon asks whether the AI generated the content or helped you improve your own. Both are asking Janis’s question in more words.

If you’re the source and the editor, most of the problems in this article don’t apply to you. If you’re neither, none of the workarounds will help.

On pride, which is a different question entirely

The other answer in that thread was blunter: it’s not you, it’s the AI that should take the pride.

This one isn’t about law or platforms, and it shouldn’t be answered as though it were. It’s about what you were doing this for.

I think the honest version goes like this. Nobody believes a carpenter should credit the saw. But if the saw designed the house, it isn’t the carpenter’s house — and no amount of sanding the doorframes makes it so. Pride tracks decisions. If you made the decisions about what this book is, what it’s about, which scene earns its place and which sentence is doing the work, then the satisfaction is yours and you needn’t feel odd about the tools. If the decisions were made for you and you tidied the result, some part of you will know. That feeling is information, not neurosis.

Which is why I’d gently push back on framing this as AI should take the pride. The tool isn’t taking anything. It has no stake in it. The question is only ever what the writer did — and that’s a question the writer can answer honestly in about thirty seconds, without anyone else adjudicating.

What the critics have right

I use these tools and I’ll still say this plainly, because pretending otherwise is how you lose the argument with people who can see it happening.

The flood is real. There is now an enormous volume of generated books competing for attention, and it’s making life materially harder for people who write slowly and carefully. It clogs discoverability. It buries good midlist work. It makes readers suspicious by default, which means writers who did the work now have to prove they did — a cost imposed on them by people they’ve never met.

The hollowness is real too, in a specific way. Writing is thinking. Prose that was generated rather than worked out tends to read smoothly and collapse under pressure: ask why a character made a choice, or what a scene is structurally doing, and there’s no answer, because nobody worked it out. I can usually tell within a page. Most editors can.

None of that is answered by “but my book is good.” It’s the cost your good book is participating in, which is a fair thing for a writing group to be angry about.

Where I land

“What’s the problem if the book is good” is a reasonable question with an unreasonable premise — that goodness settles everything else.

It doesn’t. You can write a good book you don’t own. You can write a good book you’re obliged to disclose. You can write a good book and find you didn’t get the thing you actually started writing for. Those are three different problems and none of them is answered by the manuscript being competent.

And the reverse holds. If you were the source and the editor — if the thinking was yours and the judgment was yours — then the tools you used to get there are about as interesting as which word processor you opened. That’s not a defence of AI. It’s just where the line has always been, for every tool writers have ever argued about.

The argument worth having isn’t AI: yes or no. It’s who made the decisions. Everyone in that thread, on both sides, was already arguing about that. They just hadn’t noticed they agreed on the test.

Sources: the human authorship requirement discussed here reflects the Supreme Court’s March 2026 denial of certiorari in Thaler v. Perlmutter and the US Copyright Office’s report on the copyrightability of AI-generated material. Amazon’s definitions are from the KDP Content Guidelines. This is background, not legal advice.