You Still Wrote It. AI Just Helps You See It Clearly.

The Chapter-by-Chapter (CbC) editorial protocol is an original methodology developed and owned by veteran author Craig A. Shuey.

What drove me here wasn't inspiration. It was exhaustion with tools that kept falling short, chapter after chapter.

I tried the obvious approach first: hand a manuscript to an AI, ask it to help, trust that "help" meant what I thought it meant. I started with Copilot for M365 — it was already sitting inside Word, which felt like the natural place to pilot this — then branched out through ChatGPT, Perplexity, Gemini, and Claude direct. One full-length novel, hundreds of pages of continuity and voice and lore I'd spent years building, run through all five, looking for something that wouldn't buckle under its own weight.

Every one of them buckled. Not occasionally. Reliably, and in the same handful of ways, over and over, just wearing different names depending on which platform I was cursing at that week:

• Context limitations. The manuscript simply didn't fit. Something always slid out of view — a rule, a name, a detail I'd locked in three chapters back — gone, without so much as a flicker to warn me.

• Drift. A rule the AI followed faithfully in chapter three would be quietly, confidently abandoned by chapter fifteen. No announcement. I'd only find out when the damage was already sitting on the page.

• Hallucination. The worst kind of wrong — stated with total confidence, dressed up as fact. A locked term used incorrectly. A plot detail invented outright and delivered like something I'd written myself.

• Amnesia. Every new session, the slate wiped clean. Whatever the AI seemed to understand about my manuscript's rules the day before was simply gone, and I was rebuilding it from scratch, by hand, again.

• Upload/download and token overhead. The same files, re-attached, session after session. Tokens spent re-explaining context that should never have needed re-explaining. Skill invocations piling up for edits that should have taken seconds.

None of that is a knock on any one platform — it's what happens when you point a general-purpose tool at a problem it was never built to hold: long-form, continuity-dependent, high-stakes creative work, at scale. I gave all five a real chance. Claude was the one where I could actually build something around those failures and have it hold — real session-startup discipline, state that lived in files instead of a fading conversational memory, a logging structure the AI would respect instead of quietly wriggling out from under. That's the platform I built CbC in. But the methodology itself belongs to no one's model. It's a response to failures every AI editing tool shares — solved with enough structure that those failures simply can't reach your manuscript anymore.

One distinction is worth being upfront about. CbC is built around Claude operating on your actual files — reading and writing directly against a real folder on disk — not the plain chat window where you paste text back and forth. That's Claude's computer-use mode (Cowork, in the desktop app), not standard Chat. Standard Chat can run a version of CbC, but you're back to manually uploading and downloading files every session — exactly the overhead problem above, reintroduced. The file-based version is where CbC actually stops fighting the platform, and it isn't available on Claude's free tier — it requires a paid subscription. If you're serious about using this on a full manuscript, budget for that going in.

That's where CbC came from. Not a feature I wanted. A wall I kept hitting, until I got tired enough to build a door.

What came out the other side

Once each failure had a name, I could design directly against it — and the fixes for "the AI keeps hallucinating and forgetting" turned out to be exactly the fixes authors need for the fears that keep them from trying AI editing in the first place.

"It'll steal my work." The fix for hallucination and drift was blunt: stop letting the AI generate anything at all. Confine it to flagging problems in text that already exists — never inventing so much as a clause of its own. That single restriction, built purely out of self-defense, turned out to be the whole answer to IP anxiety too. Nothing enters your manuscript that wasn't already yours, because the tool was never handed the authority to add anything in the first place. One fix. Two fears, gone.

"I'll lose control — or credit." The fix for amnesia was to make every session start cold from the files, never from memory, and to require an explicit, logged ruling on every single item before anything touched the manuscript. No inferred approval. No quiet bundling of ten small decisions into one big one. I didn't build that to protect your authorial control as some noble principle — I built it because it was the only way to stop an amnesiac AI from overstepping the moment my back was turned. It happens to also be the exact structure that keeps you the author of every decision in your book.

"I don't trust it to be reliable." This is the fear hallucination makes unavoidable, and rightly so. A tool that can state something false with total, cheerful confidence has no business being taken at its word — not once, not ever. So CbC doesn't ask you to. A phase only counts as "complete" if the log can actually show the items reviewed to earn that label. A checkmark with nothing behind it isn't a pass. It's a red flag. I wrote that rule down because its absence burned me — more than once — before I finally stopped trusting the checkmark and started demanding the receipt.

From "the fix I needed" to a protocol anyone can use

CbC wasn't built as a product. It was built because my manuscript needed to survive being edited by something that could hallucinate, drift, and forget — in that order, sometimes in the same session. But none of that is unique to my book. It's what waits for any author or editor the moment a manuscript grows long enough, and complex enough, for an AI's native limitations to stop being theoretical and start showing up on the page.

That's why this is becoming a course, not just a war story. The CbC Protocol: Applying Chapter-by-Chapter Editing to Any Manuscript walks authors and editors through adapting the full methodology — the five diagnostic phases, the standing continuity checks, the periodic audit, the file-and-logging discipline that makes all of it trustworthy — to their own project, regardless of genre, length, or platform. It's a paid course, and attendees who complete it receive a certificate.

If the fears above are the reason you haven't tried AI editing yet, that's exactly who this was built for.

Details on the course, enrollment, and certification are coming soon.