Amazon KDP's AI rules,
translated for novelists.

Since 2023, Kindle Direct Publishing has required authors to declare AI-generated content at publishing time. The policy is short. The confusion it produces is not — because the line Amazon draws is not the line most writers expect.

The two categories that decide everything

KDP sorts AI involvement into exactly two buckets, and the entire disclosure question turns on which one your book falls into:

• AI-generated — the tool produced the words (or images, or translation). Amazon is explicit that this stays AI-generated even if you substantially edited the output afterward. Generated-then-revised is still generated.
• AI-assisted — you produced the text yourself and used AI to edit, refine, error-check, brainstorm, or outline. This does not require disclosure.

Notice what the distinction ignores: quality, percentage, and effort. A chapter the model drafted and you spent ten hours rewriting is disclosable. A chapter you wrote at 2 a.m. and ran through an AI line-edit is not. The question is never how much work did I do — it's who put the first version of these sentences on the page.

The gray zone every hybrid writer lives in

Real drafting is messier than two buckets. You wrote the scene, but the model suggested three sentences of description you kept. You dictated the plot beats and the tool expanded them. You translated your own paragraph through an AI and back. Modern manuscripts are often mosaics — mostly human, with machine-made tiles scattered through.

For mosaic manuscripts the practical questions become: which passages actually read as machine-generated, and how much of the book do they cover? That's an auditing problem, and it's the reason we built a detector specifically for fiction — one that names the pattern and quotes the passage for every flag, instead of handing you an unexplainable percentage. Knowing where the AI-flavored prose sits lets you make the disclosure call deliberately — or rewrite those passages until the question disappears.

What Amazon does with the disclosure

As of this writing, the declaration is between you and Amazon: it does not appear on your product page, and Amazon hasn't announced ranking penalties tied to it. Two cautions belong next to that sentence. First, platform policies change without much notice — check the current KDP content guidelines before you publish. Second, Amazon reserves the right to remove content that creates a poor customer experience, and AI-heavy prose increasingly generates exactly the reviews that trigger that scrutiny.

The risk nobody discloses: readers

The disclosure checkbox is the smaller half of the problem. The larger half is that romance and fantasy readers — the most voracious buyers on the platform — have become alarmingly good at spotting machine prose, and they put their findings in one-star reviews. An AI slop accusation in your review section does more commercial damage than any disclosure ever will, and it lands on fully human books too when the prose happens to lean on the constructions AI made infamous.

That's the case for auditing before publishing regardless of how you drafted: find the passages that read machine-made — whatever their true origin — and fix them while they're still yours to fix. The patterns are knowable; we've catalogued the seven most common tells in AI fiction, and our audits are calibrated so human styles don't get punished: all four of our public-domain calibration novels score 0 out of 100.

A sane pre-publish checklist

1. Classify honestly. Did a tool write any of the prose? If yes, you're disclosing — editing it afterward doesn't change the category.
2. Audit the manuscript. Run it through a fiction-calibrated detector and read every flagged passage. You're looking for both genuine AI leftovers and human passages that could be mistaken for them.
3. Rewrite what reads wrong. Not with a so-called humanizer tool — those don't move structural audits — but by hand, where your voice is strongest.
4. Keep your drafts. Version history is the provenance trail that settles accusations quickly.

Frequently asked questions