Do I have to declare AI use when I publish?
On Amazon KDP you are asked to disclose AI-generated content at upload, while AI-assisted content does not require disclosure. No detector can decide which applies: answer from what actually happened in your writing process and recheck the platform's current wording when you publish.
Assisted or generated? The line, in practice
Platform definitions turn on who created the content. These are worked examples of the distinction as the major platforms currently frame it — always check the current wording where you are publishing.
| What you did | How it is normally characterised | Why |
|---|---|---|
| Wrote it yourself, used a tool to check spelling and grammar | Assisted | You created the content. This has never been controversial — it is a spell-checker. |
| Dictated it, had it transcribed, edited the transcript | Assisted | The words are yours. Transcription is a recording of your speech. |
| Wrote it, asked a tool to cut repetition without changing wording | Assisted | Editing your own text, and the wording remains yours. |
| Wrote it, asked a tool to suggest a better structure, then rewrote | Assisted | Advice acted on by you is not generated content. |
| Asked a tool to write a chapter from a prompt, then edited it | Generated | The tool created the content. Editing afterwards does not change the origin. |
| Generated a cover image | Generated | Image generation is explicitly covered by KDP's disclosure question. |
| Used a tool to translate the book | Generated | Translation is named specifically in the disclosure categories. |
Try a separate free writing review
A separate writing review for clarity and reader impact; it does not check AI use, prove authorship or decide disclosure. Send up to 5,000 words. You get back what the draft appears to be doing to somebody who wasn't there when you wrote it, the strongest thing in it named, the exact place a reader starts to drift, and three revision priorities in order.
A separate writing review for clarity and reader impact; it does not check AI use, prove authorship or decide disclosure. An AI-assisted read of your own pages. Not a human edit, not a line edit, and not a scored report.
Where the requirements currently sit
Amazon KDP asks, at upload, whether the book contains AI-generated text, images or translations, and asks you to disclose if so. AI-assisted content — where you created the work and used tools to refine it — is not required to be disclosed. The disclosure is not published on the book's page; it goes to Amazon.
Other retailers and aggregators vary, and several have added their own questions in the last two years.
Competitions, awards and literary agents frequently have stricter rules than platforms, and some prohibit AI involvement altogether. Read the specific rules; do not assume the platform standard applies.
Traditional publishers are writing AI clauses into contracts, and what they require is negotiated rather than standard.
This area is moving. Check the current wording at the point of publication — that is not a disclaimer, it is the only sensible practice.
Why the distinction is real
It is tempting to treat 'assisted versus generated' as a fudge. It is not: it maps onto a genuine difference in where the content came from.
If you dictated forty hours about your mother's life, had it transcribed, and edited it into chapters, every idea and nearly every sentence originated with you. If you typed write a chapter about a woman in 1950s Kerry and edited the result, they did not.
Both are legitimate ways to make a book, and people will differ on the second. But they are different things, and a disclosure regime that could not distinguish them would be useless.
Detectors, and why you should not trust them
AI-detection tools are unreliable in both directions, and this matters practically because people are being accused on the strength of them.
They produce false positives on non-native English, on formal or academic registers, and on heavily edited prose — because clean, consistent, low-variance writing is what they score as machine-like. Careful editing makes your writing look more generated, which is a perverse and well-documented outcome.
They produce false negatives on lightly-edited generated text.
So: do not use a detector score as evidence of anything, do not panic at a high score on your own writing, and be sceptical of anyone who does either.
What to do if you are asked to prove it is yours
This comes up with competitions, agents and occasionally readers, and the answer is provenance rather than a detector score.
Keep your drafts, dated. Keep the recordings if you dictated. Keep your notebooks, your notes, your version history. A folder of eleven dated drafts with visible development is far more persuasive than any tool's opinion.
This is worth doing prospectively rather than after an accusation, and it costs nothing.
About Your Book Pro
Your Book Pro is a book-writing platform built in Britain. Start with an idea, notes, half a draft, a podcast, or a finished manuscript, and use our purpose-built AI tools for planning, feedback, editing, proofreading, cover designs, and to create print-ready files. Writing is totally free: you can type, upload or speak your book as much as you like, and use only the editing tools you need with Plus, or go unlimited with Pro.
Produce world-class drafts at the fraction of the cost of hiring a human editorial team.
Every account starts with 1,000 welcome credits, and a plain export of your own words is free.
Each guide says plainly when a person does the job better than AI. Guides on legal topics are general information, not legal advice.
AI Footprint Checker reads the current whole manuscript in parts and flags passages whose style may strike an editor or reader as generic or machine-like. It is an advisory writing instrument, not a detector verdict. It cannot know who wrote the words, measure how much was actually generated, or decide what you must disclose.
Use it to find prose worth reviewing in your own voice. Answer disclosure questions from what actually happened in your writing process, and check the current requirements where you publish.
This is general information about published platform policies, not legal advice.
Run the advisory check on your book
Create or sign in, open your book, then choose AI Footprint on Prepare. Credits are required on Plus and included in Pro.
Create or sign inQuestions people ask
- Does Amazon KDP require AI disclosure?
- KDP asks at upload whether the book contains AI-generated text, images or translations. AI-assisted content, where you created the work and used tools to refine it, is not required to be disclosed. Check the current wording when you publish.
- Is dictating my book AI use?
- Transcribing your own speech is assisted, not generated — the words are yours. That is the same category as using a spell-checker, and it is how a great many books have been made for decades.
- Will readers know?
- Detectors are unreliable in both directions, so no, not reliably. Readers do notice blandness, though, which is the real risk and a writing problem rather than a disclosure one.
- Do competitions allow AI use?
- Many restrict it and some prohibit it entirely, with rules stricter than any platform's. Read the specific competition rules — do not assume.
- What if I'm wrongly accused?
- Show provenance: dated drafts, recordings, notes, version history. A visible development trail is far more convincing than arguing with a detector score.