Can you use AI to edit my book?

Yes, you can use AI to edit your book. The detail behind that yes comes down to one line: who made the words.

Amazon KDP sorts this into two buckets. If you wrote the content yourself and then used AI tools to edit it, refine it, check it for errors or otherwise improve it, KDP calls that “AI-assisted”, and you don’t have to tell KDP about it. That’s the bit most people want to hear.

The other bucket is “AI-generated”. If an AI tool created the actual content (text, images or translations), it counts as AI-generated, even if you applied substantial edits afterwards, and KDP requires you to declare it when you publish or republish. So the question to ask yourself is simple: whose sentences are these?

How to do it well

Think of editing as passes rather than one big go. Start with structure: pacing, plot holes, repetition. The chapter that sags in the middle, the clue you planted and then forgot about, the same striking phrase turning up four times. An AI can flag those if you ask it to look for them, and none of that involves it writing your book.

Then line editing, which is where the risk of drifting creeps in. Last comes proofreading: typos, consistency of names and timelines, and spelling variants. If you’re writing for UK readers that last one matters more than you’d think, because a tool will slide into American spellings unless you tell it not to.

Ask for suggestions with reasons. If the tool hands you a polished chapter, you can’t see what changed, and your voice gets quietly sanded off. If it lists each change with a reason, you can accept them one by one, keep the ones that sound like you and reject the rest.

Paste a chapter at a time. A chapter is a manageable chunk to review properly. It also helps to keep a style sheet (names, spellings, capitalisation) and give it to the tool at the start, so it isn’t guessing whether it’s “Mum” or “Mom”, or whether your detective’s surname has an e on the end.

Here’s a prompt that does what you want and nothing more:

Proofread this chapter in British English. List each error with the original sentence and a suggested fix. Do not rewrite style or change my wording beyond the fix.

That last sentence keeps the tool in the proofreader’s chair and out of the author’s.

One more, unglamorous point. AI won’t catch everything, and it can introduce errors of its own. A final human read still matters, whether that’s you (ideally after a few days away from the manuscript) or a professional editor. If you’re still at the drafting stage, see can you use AI to write a book?

Rules and risks

The KDP line. If the AI rewrites whole passages and you keep its wording, you’re moving towards AI-generated text, and that’s where you’d need to declare it. Translations too: AI translations count as AI-generated. Using a tool to catch a misplaced comma is one thing; accepting three paragraphs of its prose is another. KDP also says you’re responsible for verifying that AI-assisted content follows all its content guidelines, including intellectual property rights. The checking falls on you.

Privacy. This is the one people skip, and it matters for an unpublished manuscript. Google’s Gemini Apps privacy hub, as one example, says human reviewers (including trained reviewers from service providers) review some data, and asks people not to enter confidential information they wouldn’t want a reviewer to see, or Google to use to improve its services. Files you share count as part of the data collected. Temporary chats aren’t used to train Google’s AI models, and temporary chats and chats with Keep activity off are kept with the account for 72 hours.

The UK National Cyber Security Centre (NCSC) is blunter. It says queries to a public AI chatbot are visible to the provider, are stored, and will almost certainly be used for developing the service or model at some point. Its advice is to keep sensitive information out of them and to read the terms of use and privacy policy first. A novel nobody has read yet is, for most writers, sensitive. Check the settings before you upload anything, and use temporary or no-history modes where they’re offered.

Errors. The Department for Education points out that generative AI can produce inaccurate or false information presented as fact. That applies to edits too. An AI can change the meaning of a sentence while “fixing” it, or introduce a mistake where there wasn’t one. A tidy-looking suggestion isn’t necessarily a correct one, which is why accepting changes one at a time beats “accept all”.

This is general information, not legal advice.