- Publishers were among the first AI data sellers: Wiley expected $44 million from AI rights deals and Shutterstock made $104 million in a year.
- Reddit's deal with Google was reported at about $60 million a year.
- The rights question is whether contributor, author and freelancer agreements let you license for AI training.
- Court rulings against pirated sources have made licensed archives more valuable.
What publishers have earned
| Licensor | Reported value | Source |
|---|---|---|
| Shutterstock | $104 million AI licensing revenue in a year | PetaPixel |
| About $60 million a year from Google | The Decoder | |
| Wiley | $44 million expected from AI rights deals | The Bookseller |
See more on AI data licensing deals.
Why licensed archives gained value
In Bartz v. Anthropic, the court found training on lawfully bought books could be fair use, but using pirated copies to build a library was not, and the case settled for $1.5 billion. Clean, licensed sources reduce that risk for buyers. See the AI copyright lawsuits tracker.
Rights to check
- Author and contributor contracts. Do they grant you the right to license for AI training, or only to publish? Wiley's deals drew attention because authors had no opt-out (The Bookseller).
- Freelancers and agencies. Commissioned work may be licensed to you for limited uses only.
- Third-party content inside your archive: images, quotes, syndicated material.
- User-generated content. Community platforms need terms that cover licensing.
- Personal information in comments, letters and photos.
Structuring the deal
- Define the corpus precisely: titles, date ranges, formats.
- Decide on training only, or training plus display or retrieval (which is often priced separately).
- Set term, exclusivity and what happens to trained models at the end.
- Consider sharing proceeds with contributors to support future licensing.
Published content and crawling
If your content is public, AI crawlers may already have collected it. Blocking crawlers and offering paid access are two sides of the same strategy; Human Native, now part of Cloudflare, is building tools for that. Private and unpublished archives, such as backlists not online, raw footage and unpublished manuscripts, are where direct licensing has the most leverage. See books and long-form text.
Read next
Questions
Do publishers need author permission to license books for AI training?
It depends on the contract. Some grant broad rights that cover AI licensing; others do not. Check each contract or seek consent.
Is licensing for training different from licensing for display?
Yes. Training uses content to build a model; display or retrieval shows content to users. They are usually priced and licensed separately.