AI Wants to Inhale My Montana Book for $340

Should You Sell Your Book to AI for $340? A Practical Guide for Writers

A tech company shows up with a "one-time" $340 offer to train their model on your book. Do you take it? Or does that tiny check become a permission slip to clone your voice and undercut your future work?

Here's a clear way to evaluate the trade, protect your rights, and decide with a cool head.

What "AI training" actually means

Large language models don't read your book for meaning. They digest your sentences as data to predict the next word. That process can surface your phrasing, structure, and ideas in outputs-sometimes close to verbatim if guardrails fail.

Treat this as a rights and market decision, not a philosophical one. The question is: what are you licensing, for how long, and at what price?

The real risks to weigh

• Sales cannibalization: If the model learns your style and facts, it can answer questions your book used to answer.
• Undervaluation: A small "one-time" fee can set a precedent you'll regret across your catalog.
• Lack of control: Many companies won't disclose model details, data retention, or safety against regurgitation.
• Piracy baseline: Your work may already be in gray-market datasets (e.g., library mirrors and scrapes). That's a reason to get paid and set boundaries, not to give up your leverage.

A decision framework you can use today

Start with your numbers:

• Annual royalties for this title (last 12 months)
• Average monthly sell-through (print + ebook + audio)
• Backlist value: Does this book feed speaking, workshops, or a series?

Quick rule of thumb:

• Perpetual, all-model training rights: Ask 10-20x last year's royalties.
• 3-year, single-model license with deletion on expiry: 3-5x last year's royalties.
• Per-chapter excerpts or summaries only: 1-2x last year's royalties.

If the offer is under these floors, you're likely underpricing the risk.

Questions to ask before you say yes (or even negotiate)

• Who is the buyer and which specific model(s) will use the text?
• Scope: Training, fine-tuning, or retrieval-augmented inference only?
• Retention: How long will copies exist? Is deletion guaranteed and auditable?
• Regurgitation safety: What safeguards block close reproduction of passages?
• Attribution: Will outputs cite the book or link to purchase?
• Use limits: Exclusion from education-cheating tools or direct book summaries?
• Audit and transparency: Can you see where, when, and how your text was used?
• Indemnity: Are you covered if the model outputs infringing content derived from your text?
• MFN: Do you get "most favored nation" treatment if others are paid more?

Contract terms to insist on

• License: Non-exclusive, non-transferable, limited to named model(s) and version(s).
• Data handling: No sharing to third parties; encrypted storage; prompt deletion on request or at term end.
• Output controls: No outputs that reproduce "substantially similar" passages; filter checks required.
• Market protection: No use that competes with the primary market of the book (full summaries, study guides, or "book in a box").
• Reporting: Quarterly reports on use and safety performance.
• Payment: Upfront fee plus either per-user or per-token variable fee, with a guaranteed annual minimum.
• Audit rights: Independent verification of deletion and scope compliance.

Pricing options that beat a flat check

• Time-bound license: 24-36 months with mandatory deletion and renewal at your option.
• Tiered scope: Higher fee for whole-book ingestion; lower for excerpts or metadata only.
• Hybrid: Upfront fee (floor) plus revenue share tied to model usage.
• Catalog bundle: One agreement covering multiple backlist titles with per-title reporting.

If you decline

• Register your copyrights for all titles not yet registered.
• Use access controls for digital files; watermark PDFs supplied to reviewers or partners.
• Monitor for regurgitation by testing leading models with unique phrases from your work.
• Send takedowns for pirated copies and dataset mirrors.
• Join collective efforts and stay current on policy updates via the U.S. Copyright Office: AI Initiative.

If you accept

• Keep the license narrow: specific models, specific versions, clear term, deletion on expiry.
• Secure upside: upfront fee + variable component + MFN.
• Protect your market: attribution links and exclusions for study guides or book-replacement outputs.
• Plan the release: Announce the deal on your terms and direct readers to buy links.

How to stay valuable as AI spreads

• Write from lived access: on-the-ground reporting, interviews, field notes, and proprietary data.
• Productize your voice: newsletters, workshops, communities, and premium archives.
• Use AI as an assistant for outlines, fact checks, and edits-keep the final prose yours.
• Build a moat: distinct POV, recurring frameworks, and formats that reward paying attention.

AI predicts words. Writers create meaning. Price your meaning with intent, license with limits, and keep building work that a model can't fake.

Useful resources

• U.S. Copyright Office: AI and Copyright
• AI tools overview for copywriting (use responsibly)