Introduction (recap)
In Part 1 of this series, Using AI as scaffolding, not a shortcut, I described how I gave myself five weeks to write, edit, and publish my first book — and why I turned that constraint into an experiment.
If you’re curious how the writing itself came together, I’d start there.
What follows is where the real work began: editing.
The part I underestimated
Before I started The Doodle Principle — How AI Becomes Your Partner in Curiosity and Creativity, I assumed editing would be a linear cleanup step. Fix grammar. Tighten sentences. Maybe catch a few mistakes I missed.
That assumption was wrong.
Editing isn’t one task. It’s several fundamentally different kinds of thinking — structural coherence, factual accuracy, voice, rhythm, tone, consistency.
Trying to do all of that at once is exhausting, and it’s one of the reasons I believe first-time authors trying to do all the steps themselves stall, burn out, or never publish at all.
Without AI, my options were limited:
Spend months rereading, revising, second-guessing, losing momentum, and getting frustrated
Or hire a professional editor I didn’t have the budget for
Instead, I asked AI to help me answer a simple question:
What’s the right way to edit this without losing my voice or my sanity?
The answer surprised me.
Editing in passes, not all at once
Rather than one monolithic edit, AI recommended three distinct passes, each with a different purpose — and, critically, a different AI partner based on how each one tended to think.
At first, this felt excessive. Why not upload the manuscript once and be done?
Because editing isn’t a single skill.
It’s a sequence.
And mixing those skills together creates drift, overcorrection, and tone loss.
Editing, I came to realize, is a writer’s scaffolding — the temporary structure that helps ensure the author’s best work is actually visible. AI became my editing scaffolding.
Pass 1: Structural editing with ChatGPT
The first pass wasn’t about sentences at all. It was about thinking.
I used ChatGPT as a structural or developmental editor, working chapter by chapter rather than on the full manuscript at once. That decision mattered. Uploading everything at once increases the risk of stylistic drift and overgeneralization.
This pass focused on questions like:
Does this chapter deliver on its intent?
Are ideas introduced in the right order?
Where does the argument wander or repeat itself?
What assumptions am I making that aren’t explicit?
In traditional publishing, this would be called a developmental edit. With AI, it felt more like a thinking partner that never got tired of asking, “Does this actually work?”
ChatGPT consistently behaved like a generalist. It wanted to zoom out, reorganize ideas, surface assumptions, and test whether a chapter actually delivered on what it claimed. That tendency made it well suited for structural work, where breadth of reasoning mattered more than sentence-level polish.
What it didn’t do was rewrite the book. I rejected plenty of suggestions. Others I accepted not because they were perfect, but because they clarified something I was already trying to say.
This pass set the foundation. Without it, everything that followed would have been cosmetic.
Pass 2: Content and clarity checking with Gemini
Once the structure held together, I moved to clarity and accuracy.
This is where Gemini came in.
I used it as a content and clarity editor, focusing on:
Explanations that felt muddy or overly compressed
Places where I assumed too much reader knowledge
Factual consistency and light fact-checking
Sentences that made sense to me but might not land for others
This wasn’t about style. It was about comprehension.
Gemini had a tendency to pause on ambiguity. It regularly flagged places where something was technically correct but open to misinterpretation, or where a reader could reasonably challenge a claim based on wording alone. That made it particularly effective for clarity checks and factual accuracy, even when it suggested very few changes.
In practice, many chapters came back nearly clean. But when Gemini did raise a concern, it was usually worth addressing. In a few cases, those flags sent me back to my structural editor to rework tone or framing before moving forward.
Again, this was done chapter by chapter, not all at once.
The goal wasn’t perfection. It was clarity and confidence.
Pass 3: Line editing with Claude
Only after the structure and clarity were solid did I move to line editing.
This is where Claude excelled.
Claude became my sentence-level editor, focused on:
Rhythm and cadence
Word choice
Reducing unintentional repetition
Making sentences read cleanly and naturally
Claude seemed especially attuned to cohesion across longer passages. It noticed when clauses piled up, when rhythm flattened, or when a paragraph subtly drifted out of sync with the surrounding text. That made it particularly effective for line editing, where consistency of voice and flow mattered most.
This is the pass most people think of when they hear “editing,” but doing it earlier would have been a mistake. Polishing sentences before you know they belong is wasted effort.
Claude didn’t change my voice. It sharpened it.
Reading the edited chapters back was the first time I felt real confidence in the manuscript — not because it was perfect, but because it finally sounded like me on a good day.
The tedious part (and why it was still worth it)
This process wasn’t magical or instant.
It meant:
Breaking the manuscript into individual chapter files
Copying text between tools
Reviewing edits carefully
Accepting some changes and rejecting others
Revisiting sections I thought I was done with
It was slow. At times, tedious. I still felt frustrated along the way.
But compared to months of solo editing — or never publishing at all — it was a trade I’d make again without hesitation.
What surprised me most
The biggest surprise wasn’t speed. It took nearly as long to edit the book as it did to write the first draft and begin the formal editing process.
It was confidence.
By the time I finished these passes, I enjoyed reading my own work. I trusted it. I could see its flaws clearly — and that made them less intimidating, not more.
AI didn’t eliminate judgment. It required more of it.
But it made judgment sustainable.
A quick clarification
I didn’t use one AI for everything on purpose.
Each model leaned naturally toward a different kind of thinking — generalist reasoning, detail sensitivity, and long-form cohesion. Treating them as interchangeable would have flattened the result. Treating them as specialists preserved both quality and voice.
Editing didn’t save me the most time because it was faster.
It saved me the most time because it made finishing possible.
What comes next
I assumed publishing would be the easy part.
I was wrong again.
In Part 3, I’ll describe why formatting, tooling, and distribution turned out to be the most frustrating phase — and the one where AI helped the least.
That’s where this experiment took its sharpest turn.
The ideas and concepts in this article are the author’s own. AI assisted with ideation and editing.
It’s so brave of you to shed the light on the process of writing with AI. Thank you. This is a very educating and mature article that brings lots of clarity, and I hope that people who ardently speak up against AI in writing will read this.