The gold-rush version of “AI + KDP” is simple: generate 100 books, upload them all, get rich. It’s also the fastest way to get your account flagged and your time wasted. Here’s how I’d actually start a book project with AI — one that adds a real book to the store, not more junk to the pile.
Start with a reader, not a prompt
Here’s the truth that no “publish 500 books” course wants to lead with: AI multiplies skill, but it does not invent a customer. Before you write a single word, you need a real reader with a real problem and some signal that they’re buying solutions to it.
So do the boring, valuable work first. Pick a specific niche and audience. Check whether books in that space are actually selling and whether there’s an angle that isn’t already saturated. AI is fantastic for speeding up this research — clustering subtopics, analyzing what existing books cover and miss — but the judgment call on “is this worth writing?” is yours. Skip this step and the best AI on earth just helps you write something nobody wants faster.
Use AI as a co-author, not a ghostwriter
Once you’ve got a validated idea, here’s the healthy division of labor. AI is a brilliant co-author for structure and speed — and a terrible sole author for anything you’d put your name on.
- Let AI handle: outlining the book, drafting sections from your notes, suggesting chapter structures, tightening clunky paragraphs, generating exercises or summaries.
- You must own: the actual expertise, the real examples, the voice, and — critically — fact-checking every claim it makes.
Think of it as a fast writing assistant who’s confident, tireless, and occasionally makes things up. Great for momentum. Never the final word. A book that’s 100% raw AI output reads like it, and readers leave reviews that say exactly that.
Know Amazon’s actual rules (they changed)
This is where the “flood the store” crowd gets burned, so pay attention. Amazon has tightened up specifically because of AI junk:
- Disclosure: when you publish, Amazon asks you to disclose AI-generated content. Answer honestly — there’s a meaningful difference between “AI-assisted” (you edited and directed it) and “AI-generated” (it wrote it), and they treat them differently.
- Volume limits: there’s a cap of roughly three new titles per day per account, introduced directly to curb mass AI uploads. The “publish 100 books this week” fantasy is dead on arrival.
- Quality enforcement: Amazon actively removes content it deems low-quality or duplicative. Thin, spammy books don’t just fail to sell — they put your whole account at risk.
The parts worth spending real effort on
If you’re only going to invest human time in two things, make them these: the writing quality and the cover. A genuinely good cover does more for your click-through and sales than another rushed title ever will — this is one place I’d happily pay a designer or spend real time in a tool like Kittl. And a book that actually delivers on its promise earns the reviews that carry it long after launch.
How I’d actually start this week
Keep it small and real. Pick one validated topic you genuinely know something about. Use AI to build the outline and accelerate the draft. Do a serious human editing pass for accuracy and voice. Commission or design a strong cover. Publish one good book, disclose honestly, and learn the whole pipeline end to end.
One real book teaches you more — and earns more — than fifty thin ones that get you flagged. Quality isn’t the slow path here. It’s the only path that still works.
Validate a KDP niche in fifteen minutes
Before you write a word, run this quick check on Amazon itself. Search your topic and look at the top books’ Best Sellers Rank — a rank under roughly 100,000 in the paid store signals real, current sales, while ranks in the millions mean nobody’s buying. Then scan the review counts and the actual reviews: lots of reviews with common complaints is gold, because it’s proven demand plus a gap you can fill. If the top results are ancient, thin, or nonexistent, the niche is either dead or genuinely open — tell which by whether anything is selling at all.
What’s the book you’d actually write if you did it right? Tell me the topic in the comments and I’ll suggest how I’d validate it first.
