Amazon flagged your book.
Now what?
The notice rarely says the word. Here is how to read it, what not to do next, and how to answer with something better than insistence.
You uploaded a manuscript you wrote yourself, and Amazon put it under review. The email is short, generic, and says almost nothing about why. Before you rewrite a single sentence, slow down — because the most common responses to this notice are the ones that make it worse.
Read what the notice actually says
Most of these messages never mention artificial intelligence. They refer to content quality, to guidelines, or simply state that a title is under review. Authors fill that silence with the worst interpretation, and then start fixing a problem nobody has confirmed they have.
Ordinary causes are far more common than AI suspicion: metadata that resembles an existing title, formatting the converter dislikes, public-domain material Amazon requires you to declare differently, or a categories-and-keywords mismatch. Rule those out first. If the notice does explicitly reference AI-generated content, then you are in disclosure territory, and it is worth re-reading how Amazon separates AI-generated from AI-assisted work — the line is narrower than most writers assume, and the answer determines what you owe them.
The three reflexes that hurt you
Rewriting in a panic. Substantially altering a manuscript that is mid-review muddies the record. If you later need to show a reviewer what you submitted and how it evolved, you have just added a large unexplained edit at the worst possible moment.
Reaching for a humanizer. These tools mostly rotate synonyms and jitter sentence lengths. That moves surface statistics a little and structural patterns almost not at all, which is why humanizers do not reliably clear a fiction-calibrated audit. They also tend to flatten the voice a human reviewer is reading for.
Appealing with feeling instead of evidence. I wrote every word myself is true and unverifiable. It reads identically whether or not it is accurate, so it moves nothing.
Answer with provenance
The strongest reply to an authorship question is a timeline. Cloud document version history showing the manuscript accreting over weeks. Dated backups. Outlines and notes. Messages to beta readers or an editor with drafts attached. None of this is exotic — most writers already have it and never think to assemble it.
That evidence does something a score cannot: it shows the work being made. We wrote a fuller walkthrough in how to prove you wrote your novel, and the short version is that a reviewer trusts a paper trail long before they trust a number.
Then find out what your prose actually looks like
Provenance answers the accusation. It does not tell you why the question arose. For that you need to know which passages in your book read as machine-patterned to a detector — and general-purpose detectors are unreliable narrators here. They were tuned on essays, where clean, plain, declarative prose is unusual. In fiction it is a style. That mismatch is why a generic perplexity detector rates Hemingway at roughly 73 percent AI, an outcome we walk through in why Hemingway gets flagged.
A fiction-calibrated audit answers a narrower and more useful question: which specific lines carry the patterns, and why. All four of the public-domain novels we calibrate against score 0 out of 100, while contemporary AI-written samples land between 16 and 23. Our full methodology lists the samples, the thresholds, and the known limitations, including where the approach can be gamed.
What to do this week
- Re-read the notice literally. Note what it says, not what you fear it means.
- Freeze the manuscript. No rewrites while a review is open.
- Assemble the timeline. Version history, dated drafts, correspondence.
- Audit the text so you know which passages invite the question, and can speak to them specifically.
- Reply once, with documents. Short, factual, and attached — then wait.
None of this guarantees an outcome; Amazon publishes neither its criteria nor its timelines, and anyone promising otherwise is guessing. What it does guarantee is that you are answering with evidence rather than adjectives — and that if the question ever comes up again, from a platform, an agent, or a reviewer, you will already know exactly what your prose looks like from the outside.
Frequently asked questions
Does Amazon actually say the word AI in the notice?
Usually not. Most authors receive language about content quality, a title that does not meet guidelines, or a book placed under review — with no mention of AI at all. That vagueness is deliberate on Amazon's side and maddening on yours. It also means you should not assume AI is the cause until you have ruled out the ordinary explanations, like formatting, metadata, or public-domain overlap.
Can Amazon prove my manuscript was written by AI?
No, and neither can anyone else. No detector on the market measures authorship. What every detector measures is statistical pattern density — how closely prose resembles the output patterns of language models. A high reading is a correlation, not a confession, which is why a reviewer's judgement and your own records matter more than any score.
Should I use a humanizer tool before I appeal?
It is the most common reflex and one of the least effective. Humanizers mostly swap synonyms and vary sentence length, which leaves the structural patterns that fiction-calibrated audits look at largely intact. Worse, if your book is already under review, rewriting it into stranger prose can make the next reader trust it less rather than more.
How long does a KDP review take?
Amazon does not publish a service level, and reported waits range from a couple of days to several weeks. Sending repeated follow-ups rarely accelerates it. The productive use of that waiting period is assembling your drafting evidence and understanding what in your prose might have triggered the flag in the first place.
I write minimalist prose. Am I more likely to be flagged?
Plain, declarative, low-adjective prose does read as statistically unremarkable to general-purpose detectors — which is exactly why Hemingway scores around 73 percent AI on a generic perplexity detector. Slopsleuth is calibrated against published fiction instead, and all four of our public-domain calibration novels score 0 out of 100.
What should I keep as proof that I wrote it?
Anything with timestamps: cloud document version history, dated backups, notes and outlines, messages to beta readers or an editor, and earlier drafts that show the manuscript changing over time. Provenance is a paper trail, and it is far more persuasive to a human reviewer than any score you can produce.
Know what your manuscript looks like before you reply.
Five fiction-calibrated audits, every flagged passage quoted, and a per-chapter breakdown you can point at. Free sample audit, no signup.
Audit your manuscript → See plans & pricing