The honest guide to publishing AI-assisted fiction.

1. Copyright

The US Copyright Office (March 2023, reaffirmed 2024) holds that purely AI-generated text is not copyrightable. A book where you wrote the prompts and accepted the AI's output verbatim has no copyright protection in the United States. A book where you substantially edited, rearranged, and rewrote the AI's output is copyrightable — but only the human-authored portions.

The UK, EU, China, and Japan have all reached similar conclusions. The practical takeaway: treat AI as a collaborator whose contributions you rewrite, not a ghostwriter whose pages you keep.

2. Training data

Every large language model was trained on a corpus that almost certainly included copyrighted books — whether they were licensed is the subject of multiple ongoing lawsuits (Authors Guild v. OpenAI, NYT v. OpenAI, etc.). You don't get to choose the training data, but you can choose tools whose providers are transparent and defending their position in court. SCRIVONA uses models from OpenAI, Anthropic, and Google, all of whom publish their data practices.

3. Disclosure

• Amazon KDP: required if AI generated content. "AI-assisted" (you edited) is exempt.
• Traditional publishers & agents: almost always required at query stage.
• Contests: most ban AI-generated work outright; check the rules.
• Readers: not legally required, but a short acknowledgements note is becoming standard practice.

4. The ethics line

The defensible position most working novelists arrive at: AI is fine for brainstorming, outlining, research, and copy-editing. AI-generated prose belongs in your draft only as raw material you rewrite. Imitating a specific living author's voice in a way that could be passed off as theirs is the bright line — don't cross it.