How we use AI in the draft, and where we refuse to

The question comes up in every discovery call. You have seen AI-generated books. You have read three pages of one and recognized the pattern: smooth sentences that say nothing specific, transitions that connect every paragraph to the next without any of them carrying weight, a complete absence of the particular incident that would make you trust the author. You want to know whether the book that comes out of this process sounds like that.

It does not. Here is the actual division.

Where AI runs in the process

The in-app coach runs on AI. It reads every section you write the moment you submit it and checks it against your locked voice file. It flags passive constructions, banned words, structural gaps between your framework pillar and the evidence you provided, and section-length drift from the five-section chapter spine. It does this in seconds. A human editor reviewing the same pass catches the same categories of error, but the human reads once a week. The coach reads every save.

The AI agent infrastructure that runs during the launch phase also handles volume work: drafting your press release in your documented voice, building your email nurture sequence from your story bank entries, mapping your twelve-week social rollout post by post. These outputs land in your inbox for a five-minute approval before anything goes out. You read them. You change what does not sound like you. The agent revises.

The system also handles podcast research. It identifies hosts in your category, pulls the episode history, drafts the pitch in your voice, and queues the outreach. Your VA sends it. Your name and your credibility go on every line.

Where we refuse to use it

The in-app coach does not write your chapters. It reads them. That distinction is the whole point. Your chapter begins with you describing a real situation from your Story Bank. Your chapter proceeds through the five-section spine in your documented sentence rhythm. Your chapter ends with an AI prompt the reader can run themselves, and you write that prompt too, because the prompt has to reflect your framework logic and only you carry that.

The editorial team does not use AI to generate voice polish notes. They read the manuscript. They write the note. The note says the thing in plain English, names the paragraph, and tells you the specific problem.

No chapter in a Crazy Simple Press book comes from a model that was handed a topic and asked to produce prose. That produces a book that reads like a category average. Category average is the competitor you are trying to make irrelevant.

Write the chapter. Use the coach to check it. That sequence is non-negotiable.