Free playbook

Publish KDP books that survive review and actually sell.

Most catalogs fail for one of two reasons: the books get taken down, or nobody finds them. Neither is a writing problem. This is the loop I run — niche, metadata, compliance, production — written for operators building a catalog, not for someone writing one novel.

I ran a KDP catalog to €400/month, then lost the account overnight to a pen name that read like a joke. Everything below is what I wish the first version of me had been handed.

The loop

Choose the niche on demand signals, not on taste

A catalog operator is not picking a passion project. You are looking for a reader who is actively trying to solve something and already buying books about it. That reader leaves evidence: they ask the same question repeatedly, and the existing answers are thin.

Look for a specific job, not a topic — 'stop my sourdough from going flat' beats 'baking'.
Check that books in that niche have recent reviews. Recent reviews mean recent buyers.
Read the 3-star reviews of the top sellers. They name the gap your book fills.
If the top of the category is dominated by strong brands with hundreds of reviews, go one level narrower.

The mistake: Choosing a niche because the keyword volume is high. High volume with strong incumbents is the hardest possible place to start.

Accept that metadata decides the ranking

The manuscript decides whether readers keep the book. The metadata decides whether they ever see it. Operators consistently underweight this because it feels like paperwork, and it is where almost all the recoverable ranking lives.

The title carries the promise; the subtitle carries the keywords. Do not make the title do both.
The description's first two lines are all most buyers read — put the outcome there, not your credentials.
The 7 backend keyword slots are phrases, not words. Use search phrases a buyer would actually type.
Category fit beats category ambition: ranking in a category you can win beats invisibility in a big one.

The mistake: Reusing the same keyword in the title, the subtitle and all 7 slots. It buys you nothing and wastes six slots.

Build the compliance layer before you need it

This is the step that ends catalogs. Amazon's rules on AI disclosure, volume and content quality are enforced by systems that do not take appeals well, and a terminated account takes the whole catalog with it — including the books that were selling.

Disclose AI use where KDP asks for it. The disclosure is not a penalty; failing to disclose is.
Use a pen name that reads like a person. A joke name is a flag, and it is the mistake that cost me an account.
Respect the daily title limit. Bursts look like exactly what the system is built to catch.
Read every book before upload. Not to enjoy it — to catch the one paragraph that gets it rejected.

The mistake: Treating volume as the strategy. Volume is a production capability. If it is also your promise, you are optimising for the behaviour Amazon is actively suppressing.

Run production as a loop, not as a launch

A catalog compounds only if each book teaches you something the next one uses. That requires publishing on a rhythm you can actually sustain and keeping a record, so you are comparing books rather than guessing.

Ship in small batches and let each batch inform the next brief.
Track per book: niche, price, category, and sales at 30 days. Four columns beat a dashboard.
Kill a niche after three books that go nowhere. Persistence in the wrong niche is the most expensive habit in publishing.
When a book works, make the next three adjacent to it. Adjacent beats novel.

The mistake: Publishing 20 books across 20 unrelated niches. You end up with 20 experiments and no signal you can act on.

Get the operator kit

The playbook is on this page, free, no signup. The kit is the working version of it — the parts you fill in.

The pre-upload review checklist (the one that keeps accounts alive)
Niche brief templates that produce a book worth selling
The metadata worksheet: title, subtitle, description, 7 keyword slots
The catalog economics sheet — what a book has to earn to be worth making

Where the tool fits

Steps 1, 3 and 4 are judgement — no tool replaces them. Step 2, plus the production itself, is mechanical, and that is the part DraftToDone.io does: one niche brief in, and the manuscript, the full wrap cover, the KDP title, subtitle, description and 7 keyword slots come out as one package, with a pen name attached. You still read the book before you upload it.

See how it works

Questions operators actually ask

Is publishing AI-assisted books allowed on Amazon KDP?

Yes, with disclosure. KDP asks you to declare AI-generated content when you upload, and it distinguishes AI-generated from AI-assisted. The rule that ends accounts is not using AI — it is failing to disclose it, ignoring the daily title limit, or uploading content that fails quality review.

How many books should a new catalog operator publish per month?

Fewer than you can technically produce. The constraint is not generation speed, it is your ability to review each book properly and to learn from the last batch. Operators who ramp slowly and keep records outperform operators who flood the account and get flagged.

What actually gets a KDP account terminated?

In practice: undisclosed AI content, exceeding the daily title limit, metadata that misleads (keyword stuffing, misused category), duplicate or near-duplicate content, and pen names that appear designed to impersonate or mislead. Any one of these can take the whole catalog, not just the offending title.

Should I use one pen name for everything or one per niche?

One per niche, and make each one read like a real person. A pen name that stays inside one subject builds a recognisable author signal for readers and for Amazon's own recommendation surfaces; one name across unrelated categories builds neither.

If you only take one thing

The catalog that wins is not the biggest one. It is the one still standing in eighteen months, because every book in it was reviewed before upload and every niche in it was chosen on evidence.

Updated 2026-07-26