2026-06-13 · all guides

AI-Generated vs AI-Assisted on KDP: Which Box Do You Check?

Why this one distinction carries the whole policy

When you publish on KDP, the setup flow asks whether your book contains AI-generated content. Everything about how you answer depends on a two-word distinction in the KDP guidelines: generated versus assisted. AI-generated content requires disclosure; AI-assisted content requires nothing at all. Get the classification right and the rest is a thirty-second checkbox exercise.

The stakes are asymmetric. Disclosing costs you nothing: the answer never appears on your product page, does not block publication, and does not flag your book to buyers. Failing to disclose when you should is a terms-of-service violation that puts your account at risk over a checkbox that was free to check honestly.

AI-generated: the AI created it, full stop

Content is AI-generated when an AI tool actually produced it: you gave a prompt and the model wrote the text, drew the image, or produced the translation. This is the category that must be disclosed, and it covers all three content types. A manuscript written by a language model is AI-generated text. A cover made with an image model is an AI-generated image. A machine-translated edition is an AI-generated translation.

The detail people most often get wrong: editing does not reclassify. If the model wrote the chapter and you spent ten hours rewriting sentences, cutting repetition, and fixing facts, it is still AI-generated under the KDP definition and still must be disclosed. The policy asks who created the content, not who finished it.

AI-assisted: you created it, the AI helped

Content is AI-assisted when you created it yourself and used AI to edit, refine, error-check, or otherwise improve it. Running your own manuscript through an AI proofreader, asking a model to suggest tighter phrasing for a paragraph you wrote, or using AI to critique your structure all fall here. So does using AI purely for ideation: brainstorming titles, outlining topics you then write up yourself, or researching questions to answer in your own words.

AI-assisted content requires no disclosure at all. You answer no to the AI-generated question and move on. Amazon drew the line this way deliberately: tools that help human authors have been normal for decades, and spell-check with better vocabulary did not need a policy.

The gray areas, called honestly

Real workflows mix both. The honest approach is to classify each piece of content by who produced its first substantive version. You outlined, the AI wrote chapters from your outline: the text is AI-generated, because the model produced the prose. The AI outlined, you wrote every chapter yourself: the text is your creation with AI assistance. You wrote a rough draft and asked the model to "rewrite this in a cleaner style": now it is genuinely ambiguous, and if the output is substantially the model’s prose rather than your sentences cleaned up, the safer honest answer is to disclose.

Note that the disclosure is per content type. A book with human-written text and an AI-generated cover discloses AI-generated images but not text. If sorting your specific workflow into the right boxes still feels murky, ebookdone maintains a free interactive checker at /tools/kdp-ai-disclosure that walks through the questions the way the KDP flow asks them.

When in doubt, disclose. There is no penalty side to a yes, and no scenario where an unnecessary disclosure hurts you. There is a real penalty side to a wrong no.

FAQ