Prove You Wrote Your
AI-Assisted Novel

Amazon can pull your book from sale if they think an AI wrote it. They don't need to be right. They just need to suspect.

When that happens, you get an email asking for "Proof of Publishing Rights." You have days to respond. If you can't prove the work is yours, they remove the book. Do it again and they can terminate your account.

You use AI. You use it honestly. You workshop characters, draft scenes, stress-test your plot. You rewrite everything three times. The book is yours. But how do you prove that to someone who's never watched you work?

AI detectors are unreliable

AI detection tools produce false positives on human-written text. They miss actual AI output that's been lightly edited. They disagree with each other. A passage flagged as "95% AI-generated" by one detector gets cleared as "100% human" by another.

And the technology is stuck in an arms race it can't win. As AI models improve, their output gets harder to distinguish from human writing. That's the whole point of better models.

If your defence is "run it through an AI detector and it passes," that's not a defence. That's a coin flip.

What actually counts as evidence

Courts and publishers look for the same things in authorship disputes: creation dates, drafts, versions, records that trace the creative process.

Not the finished manuscript. The mess that came before it. Notes, outlines, rough chapters, revisions, dead ends you abandoned, scenes you rewrote four times. That trail is hard to fake because it's boring and specific and full of wrong turns. It looks like work, because it is.

What a writing history looks like

If you write in plain text files (like Obsidian's markdown), you can use a system called git to track every change you make. Think of it as automatic version history, but much more detailed than "version 1, version 2."

At the end of every writing session, you save a snapshot. That snapshot records exactly what your manuscript looked like at that moment. Over weeks and months, those snapshots become a complete record of your book being written.

The history doesn't just prove you wrote the book. It proves you worked on it. Where you disagreed with the AI's output. Where you rewrote a scene because it didn't sound like you. Where you cut a subplot that wasn't earning its pages. That's authorship. Not typing. Deciding.

Why this matters for KDP and self-publishing

Amazon's policy distinguishes between AI-assisted content (where you direct the work and make the creative decisions) and AI-generated content (where the AI produced the output and you published it). AI-assisted content doesn't need to be disclosed on KDP. AI-generated content does.

Your writing history is the clearest possible evidence of which side of that line you're on.

What this doesn't replace

A writing history is not copyright registration. If you publish a book and someone copies it, registration with the US Copyright Office (or your country's equivalent) gives you legal standing. That costs about $65 in the US.

The writing history is the supporting evidence. The registration says "this person claims authorship." The snapshot log says "and here are months of proof."

Register your copyright when you publish. But the writing history is what you'll be glad you have if anyone ever asks how you wrote it.

How to start

• Write in plain text files. Obsidian, any markdown editor. If your files are .md on your hard drive, you're set. Word and Scrivener files can be stored but git can't see the changes inside them.
• Use Claude to handle the technical part. "Back up my work." "Push to GitHub." Plain English, every time.
• Build the habit. End of every writing session: "back up my work." One sentence. Over a few months, you'll have a detailed, timestamped record of every session.

That record might never matter. You might never get challenged. But if you do, you'll have something better than a detection score or a stack of Word files. You'll have the receipts.

Read more on Medium →