How to prove you
wrote your novel.
You can't win an argument against a percentage. You can win one with a provenance trail. Here is how to build it.
It is a strange thing to have to prove: that you wrote your own book. But it is happening to working writers now — flagged by an agent's screening tool, questioned by an editor, accused in a review. The instinct is to protest your innocence. The better move is to make the question answerable with evidence.
Why a denial doesn't work
When someone runs your manuscript through a detector and it returns a high number, your saying you wrote it does not move them — you are simply contradicting their tool. And the tool is not as reliable as it looks. Run The Sun Also Rises through a perplexity-based detector and it scores around 73 percent AI; Hemingway has been dead since 1961. We documented that case in why Hemingway gets flagged as AI, and the styles most at risk of a false positive are exactly the disciplined ones.
So the goal is not to win the percentage argument. It is to make the percentage irrelevant by producing something it cannot fake: evidence that the book was written, by you, over time.
Provenance is the whole defense
A finished manuscript looks the same whether it took a machine four seconds or took you four years. What differs is the trail behind it. A real book leaves a sediment of drafts, dead ends, and second thoughts — and that sediment is something no AI tool generates.
The evidence worth keeping, in rough order of strength:
- Cloud version history. Google Docs, Microsoft Word, and Scrivener all retain a timestamped revision trail. A document that visibly grew and changed across months is the single most persuasive artifact you have.
- Dated backups and earlier drafts. Saved versions, exported files, even messy first attempts — the rougher, the more human.
- Process documents. Outlines, research notes, character bibles, and revision letters show a mind working through the material.
- Correspondence. Beta-reader feedback, critique partner threads, and editorial notes place your book in a human timeline of conversation.
The key habit: start keeping this before you need it. The writer who can produce a year of version history the moment a question is raised ends the exchange instantly. The one who has to reconstruct it later looks defensive even when innocent.
How to respond to an accusation
If an agent, editor, or platform raises the question, the tone that works is calm and evidentiary, never wounded. Lead with the provenance: offer to share your version history and an earlier draft. Note, without heat, that perplexity-based detectors routinely misclassify published human authors — a fact, not an excuse. You are not asking them to take your word; you are handing them a way to verify.
A transparent audit of your own manuscript helps here too. Rather than a single contested number, it produces a list of specific passages and the pattern each one triggered — which moves the conversation from is this AI to here is exactly what the prose does and why. That is a discussion you can win.
See what they see, first
The strongest position is to never be surprised. Before you query or publish, audit your manuscript with a tool built for fiction that names the pattern and quotes the passage for every flag — the design philosophy behind our fiction-specific detector, whose methodology is public. Read each flagged line. Some will be genuine AI residue if you used assistance; others will be your own style, worth defending. Either way, you walk into any conversation already knowing what the other person will see.
The honest limit
One thing no one selling certainty will tell you: no detector can prove who wrote a book — ours included. A score is a statistical observation, not a verdict. That cuts both ways. It is why an accusation built on a percentage is weak, and it is why your defense should rest on something sturdier than another percentage. Provenance, plus the calm to present it, is the proof that actually holds.
Frequently asked questions
How do I prove a human wrote my book?
No detector can prove authorship, so the proof is provenance, not a score. Draft history is the strongest evidence: a manuscript that visibly evolved over weeks or months in Google Docs, Word, or Scrivener is something no AI tool produces. Keep that trail, and a single screenshot usually ends the conversation.
An agent or editor says my manuscript looks AI-generated. What do I do?
Don't argue with the percentage — respond with evidence. Share your version history, offer earlier drafts, and, politely, point out that perplexity-based detectors flag canonical human authors. Reframing the exchange from accusation to evidence is far more persuasive than insisting you are innocent.
What evidence should I keep while drafting?
Anything that shows the book taking shape over time: cloud version history, dated backups, research notes, outlines, revision letters, and beta-reader correspondence. The more your process is documented, the harder any AI accusation is to sustain. Start keeping it before you need it.
Does running my book through a detector help my case?
It helps you, first. Auditing your own manuscript with a tool that quotes the passages it flags shows you exactly what a suspicious reader will see, so you can revise the lines that read machine-made before anyone else weighs in. A transparent report also reframes a dispute around specific passages rather than a single mysterious number.
Why does my human writing get flagged at all?
Most detectors measure predictability, and disciplined prose is predictable by design — short declarative sentences, controlled vocabulary, deliberate rhythm. The craft that makes fiction clean is exactly what generic detectors mistake for a machine. Minimalist, heavily-edited, and non-native-English prose are the most common false positives.
Know what a suspicious reader will see — before they do.
Five fiction-calibrated audits. Every flag explained, every passage quoted. Free sample audit, no signup.
Try Slopsleuth free →