Disclosure: this post is written by the team behind KDP Quick Scout. This isn’t legal advice, publishing policies change, always check KDP’s own current content guidelines before you rely on anything here.
Every few weeks a version of the same question shows up in author forums: “I used ChatGPT to help outline my book, do I have to tell Amazon?” The answers underneath are usually confident and usually contradict each other. Half the thread says any AI tool at all triggers disclosure. The other half says it’s fine as long as you edited it yourself. Both are wrong in ways that matter, and the answer is sitting in Amazon’s own help pages, worded more precisely than either camp gives it credit for.
The distinction
KDP splits AI use into two categories, and only one of them needs to be disclosed.
AI-generated content is Amazon’s term for “text, images, or translations created by an AI-based tool.” This covers your manuscript text, your cover art, interior illustrations, and translated editions, if an AI tool produced the content, it’s AI-generated. Crucially, the policy specifies this applies “even if you substantially edited the AI output afterward.” Heavy editing doesn’t move AI-generated content into the undisclosed category. That’s the part the “I rewrote it so it’s fine” argument gets wrong.
AI-assisted content is everything else: brainstorming with an AI tool while you write the words yourself, using it to error-check or refine content you created, running it through a grammar checker. Amazon’s own wording here is direct: “It is not necessary to inform us of the use of such tools or processes.” That’s the part the “any AI tool means disclosure” argument gets wrong.
The line isn’t “did AI touch this book at all.” It’s “did AI produce the words or images that ended up in the book, or did it help you produce them yourself.”
Where common author workflows fall
A few concrete cases, since the abstract version of this rule is easy to agree with and still get wrong in practice:
- Asking an AI tool for chapter outline ideas, then writing every chapter yourself. Assisted. The tool never produced text that ended up in the book.
- Asking an AI tool to draft a full chapter, then heavily rewriting it. Generated, per Amazon’s own “even if substantially edited” language. The origin of the text matters more than how much you changed it afterward.
- Running your own manuscript through a grammar or style checker. Assisted. You wrote it, the tool is refining it.
- Generating a cover image with an AI art tool. Generated. Cover art is explicitly named alongside text and translations.
- Using an AI tool to research your genre, competitors, or keywords. Assisted, and arguably not even close to the line, none of that output becomes part of the book itself. This is the category a niche-research tool like KDP Quick Scout’s AI analysis falls into: it’s producing a report you read, not text that goes into your manuscript.
- Translating your own book into another language with an AI tool. Generated. Translations are named specifically, the same way text and images are.
A myth worth retiring
“If I write a detailed enough prompt, that counts as writing it myself.” It doesn’t, and this is worth being blunt about because it’s the version of the question that shows up dressed up as a technicality. Amazon’s category is about where the text came from, not about how much creative direction you gave the tool that produced it. A five-paragraph prompt that generates a chapter is still a generated chapter.
Why this is worth getting right, not just compliant
Beyond the obvious (avoiding a takedown), the disclosure categories are a genuinely useful way to think about where AI belongs in your process. Research, outlining, editing, and error-checking are exactly the parts of publishing where AI assistance is fastest and lowest-risk, you stay the book’s author, and none of it needs a disclosure checkbox. Generating the manuscript itself is a different decision with a different set of trade-offs (disclosure, quality consistency, whether readers can tell), and conflating the two categories, in either direction, just means making that decision by accident instead of on purpose.
If you’re using an AI tool anywhere in your process right now, the one question worth asking honestly is the one Amazon’s own wording already asks: did it write the words, or did you?