AI disclosure,
platform by platform.

There is no single answer to whether you must declare AI in your book, because there is no single rulebook. Each platform wrote its own policy at a different time, with a different level of teeth, and a few major stores have written nothing public at all. So the honest map is store by store.

Amazon KDP: disclose AI-generated, and edit deeply

Since 2023, KDP has asked you to declare AI-generated content on an AI-content form during publishing. The distinction it draws is the one most writers get wrong: text a tool produced counts as AI-generated even after you edit it heavily, while text you wrote yourself and merely refined with AI is AI-assisted and needs no disclosure. We unpack that split in detail in Amazon KDP's AI rules, translated for novelists.

The trap on KDP is the second filter. Even a correctly disclosed book can be caught by the disappointing-content review if the prose reads as lightly-edited machine output. The bar Amazon sets for substantial modification is high — rewriting, restructuring, and real voice work, not a polish pass. Disclose accurately, then make sure the sentences earn their place.

IngramSpark: strict wording, narrower intent

IngramSpark is the one that reads alarmingly on first contact. Its Catalog Integrity Guidelines list content created by automated means, including artificial intelligence, among prohibited material. Taken literally, that is broader than KDP's disclose-and-proceed approach.

In practice, industry commentators consistently read that clause as aimed at the flood of mass-produced, scraper-assembled titles rather than at a human-authored novel that used AI as a background tool. The wording is still broad enough that if AI touched your drafting in any meaningful way, the safe move is to ask IngramSpark directly before you upload rather than guess.

Draft2Digital: follows the disclose-at-upload model

Draft2Digital has tracked the KDP approach: declare AI-generated content during the upload flow, and human-authored books that lean on AI only as an assistant are treated as ordinary submissions. As with every platform here, accurate representation of authorship is the load-bearing requirement.

Apple Books and Kobo: the silent stores

This is where honesty matters more than a confident summary. As of mid-2026, neither Apple Books nor Kobo Writing Life publishes a clear, author-facing AI-disclosure rule the way KDP does. Silence is not permission, and it is not prohibition either — it simply means there is no checkbox to guide you.

When a platform has not written a rule, disclosing conservatively where a field exists, and keeping your own records where one does not, is the choice that ages well. Policies in this space are being rewritten constantly; the book you upload today may be governed by a rule that appears next quarter.

The risk that no platform writes down

Every policy above is about the store. The bigger exposure is the reader. Genre audiences have become strikingly good at spotting machine prose, and they record what they find in reviews — which land on fully human books too, when the writing happens to lean on the constructions AI made familiar. A disclosure checkbox is invisible to buyers; a review is not.

That is the case for auditing before you publish on any platform. Find the passages that read machine-made, whatever their real origin, and decide deliberately. Our audits are calibrated so genuine human styles are not punished — all four of our public-domain calibration novels score 0 out of 100, and the detector is built specifically for fiction rather than essays, quoting every passage it flags so you can judge each one yourself.

A platform-agnostic pre-upload routine

1. Classify your manuscript honestly. Did a tool write any of the prose? If yes, you are in disclosure territory on the stores that ask.
2. Match each store's form. Disclose AI-generated where the field exists; where a platform is silent, disclose conservatively and keep your own record.
3. Audit, then revise by hand. Read every flagged passage and rewrite the ones that genuinely lean on AI tells — in your own voice, not with a so-called humanizer tool.
4. Keep your drafts. Version history is the provenance trail that settles a dispute in one screenshot, on any platform.

The landscape is inconsistent and it will keep shifting. The constant underneath it is simple: represent your authorship accurately, and make sure your prose reads like you wrote it — because you did.

Frequently asked questions