What "reviewed, then revised" means: the AI editing pass before your book ships

Ask a chat window to write a chapter and you get exactly one thing: the first draft, delivered in a single pass, checked by nobody against the six chapters sitting next to it. That’s not a criticism of the model. It’s just what “generate” means when nothing happens after it. InkMagnet’s pipeline has a named stage for what happens after: Review & Revision, sitting between Content Generation and PDF Compilation on the progress screen you watch while a book builds. It isn’t there for show.

A first draft is not a finished chapter

Writing a full book inside a raw chat window produces prose that reads fine sentence by sentence and falls apart at the chapter level: a case study the outline promised never shows up, the closing paragraph restates the opening almost word for word, chapter six quietly redefines a term chapter two already used a different way. None of that is a hallucination in the usual sense. It’s what one-shot generation looks like when nothing checks a chapter against the plan it was supposed to follow.

Two different reviews, one pipeline

InkMagnet actually runs two separate checks, and it’s worth telling them apart. The first happens on the “Plan” step, before a single chapter is written: you approve or edit the chapter-by-chapter outline yourself, the same gate the four-step pipeline describes as catching structural drift while it’s still cheap to fix. The second happens later, on the “Review & Revision” step of the generation screen, after every chapter already has a full first draft. That one isn’t yours to approve or skip. It runs automatically, chapter by chapter, on content that already exists, which makes it a much more mechanical job than judging a five-line outline synopsis.

What the Review & Revision phase actually checks

On that second pass, an AI editor reads each chapter against the approved outline and the chapters around it, checking for the same three things a professional developmental editor’s checklist runs on: gaps where the outline promised content the draft skipped, redundant passages that repeat a point the chapter already made, and terminology that’s drifted from how an earlier chapter defined it. It fills what’s missing and trims what repeats before the chapter ever reaches PDF Compilation, and it runs on every chapter of every book, not just the ones that feel thin.

Why a second pass beats one shot, according to the research

This isn’t a marketing intuition about AI needing “a human touch.” A 2023 NeurIPS paper called Self-Refine, from Aman Madaan and coauthors, tested exactly this mechanism: the same model generates an output, then generates feedback on its own output, then revises based on that feedback, iterating without any extra training or a second model. Across seven tasks and GPT-3.5, ChatGPT and GPT-4, outputs that went through this feedback-then-refine loop were preferred by both human raters and automatic metrics over one-step generation by roughly 20 percentage points on average. A second pass that checks the first against explicit criteria measurably beats shipping the first pass raw, which is the same reason a developmental edit exists in traditional publishing and the same principle the Review & Revision phase runs on a chapter instead of a whole manuscript at once.

What this pass costs when a human does it instead

Run that same coverage-and-redundancy check by hand and you’re paying for developmental editing, which 2026 industry rate roundups put at roughly $0.03 to $0.045 per word for nonfiction, on top of whatever copyediting and proofreading cost afterward. That works out to somewhere in the low thousands of dollars for a 60,000-word manuscript, billed as a separate line item from the writing itself. It’s also the pass that gets skipped first when a self-published author is working to a tight budget: surveys of indie authors consistently find that newer, lower-earning writers default to no editing or a beta reader instead, precisely because a proper developmental pass is priced like a specialist service, not a formatting checkbox. Inside InkMagnet, the same coverage check runs on every book at every length tier, folded into the one flat price you already paid to have it researched and written: $9.99 for a 30-45 page Compact book up to $34.99 for a 161-200 page Complete one, with nothing billed separately for the editing pass.

A gap and a repeat, caught before you see either

Picture a chapter whose outline promised a competitor pricing comparison as its second section. The first draft opens strong, restates the chapter’s thesis almost verbatim in its closing paragraph, and never actually gets to the comparison it was supposed to deliver. The Review & Revision phase catches both problems in the same pass: the missing section gets written in against the original brief, and the repeated closing gets trimmed back to something that adds a new point instead of restating the old one. You never see the version with the gap, because it never reaches PDF Compilation. What lands in the WYSIWYG editor once the book is done is the version that already cleared that check. That’s also why editing the finished book later works the way it does: a scoped, chapter-level regenerate only has to handle whatever you personally want to change, because the automated pass already handled the coverage and repetition problem before your own read-through ever started.

None of this replaces your own read-through. It just means the copy you’re reading for the first time already cleared the same bar a developmental editor would have held it to, before you spent a minute on it. Start from your own topic and watch the Review & Revision phase run on the very first book you generate.