I Let AI Edit My Book. It Took Four Attempts to Get My Ideas Back.

The book I just published took four attempts. The first three weren't bad drafts in the usual sense. Page by page, each one read fine. The problem was that AI had gradually pulled them away from what I was trying to say, and I didn't notice until I was well into someone else's book.

I'm telling that story because it's the whole argument of the book, at the scale of one person.

The first attempt: one AI edits, another critiques

The first attempt was simple. I got my thoughts down and handed the editing to an AI chatbot. It worked well, at least at first. Edits came back clean, the document read consistently, and the whole process felt powerful.

Then I tried to build in a check. One chatbot did the writing. A second chatbot critiqued it, and I fed those critiques back into the first one to revise.

The critiques were good. That was the problem. Each one was sharp enough to act on, and each round of revisions moved the book a little further from where I meant it to go. Two AI systems were checking each other, and neither one knew what the book was supposed to say. Only I knew that, and I didn't supervise closely enough to hold the direction.

The second attempt: AI for positioning in the market

I asked the chatbot how my book might compare to similar titles. It handed me a completely different direction, one built for executives instead of practitioners. It sounded reasonable, so I followed it and let AI do much of the writing.

I shouldn't have. I spent days rebuilding the book around ideas that weren't mine, written in language close enough to my own that I didn't notice the drift until I was several chapters in. I got frustrated and shelved it for weeks.

The third attempt: every page read fine

For the third attempt I went back to my original manuscript. The concepts were right, but the execution was rough, so I spent weeks using AI to reorganize sections, tighten arguments and match my voice. When I finished, I felt ready to publish.

Then I ordered a printed proof and read it straight through instead of in pieces.

Every page read fine on its own. Read as a whole book, it repeated itself, lost its thread and, to put it plainly, got boring.

No single edit caused that. Each one was reasonable. The direction they added up to was not.

The fourth attempt: AI stopped setting the direction

The fourth attempt is the one you can buy. AI still helped with consistency and flow. It no longer decided what the book said. I did.

That's the distinction I learned the hard way, and it's the same one I'd ask any business to draw: AI is extremely useful for execution and dangerous as a source of judgment.

Why this matters more for a business than it did for me

AI produces plausible answers by design. Ask it a question and it answers, whether or not the answer is right, because producing a fluent, believable response is the job. It isn't trying to mislead you. It simply has no internal alarm that goes off when it's wrong.

I had one manuscript and one person checking it, and the drift still got past me three times. I caught it only because I stepped back and read the whole thing at once.

A business doesn't get that luxury. It isn't editing one book. It's making hundreds or thousands of AI-assisted decisions across sales, service, hiring, purchasing and operations, often at the same time, and usually with nobody reading the whole thing at once. Each decision can look reasonable while the overall direction moves somewhere nobody chose.

That's what I mean when I say AI rarely fails with a bang. It drifts.

Three questions to ask about your own AI use

You don't need a formal program to start noticing drift. Pick one place your business uses AI and ask:

Who decides the direction here, a person or the tool? If AI is drafting, summarizing or recommending, a named person should own the call about what it's for.
When did someone last look at the whole, not just the pieces? Reviewing outputs one at a time is exactly how I missed my own drift. Step back and look at a month of them together.
Would you notice if it started steering? If the honest answer is "probably not," that's where to start.

How this blog is written

You'd be right to ask. Each post starts as an AI draft built from my book. I edit it, check the facts against the sources, and approve it before it goes up. AI handles the execution. I decide what it says. It's the same rule I'm asking you to apply.