Per-copy licensing
Billing an AI system's access to a work by the retail copy the reader already bought. The reader purchases through the ordinary storefront; that purchase is what authorizes the model to ground its answers in the work; and the authorization is tied to the person, not to the document they made with it.
The one worked example is Google's Expert Intelligence (2026-08-27): a Play Books purchase makes a book usable inside Gemini Notebook, and a collaborator on a shared notebook is prompted to buy their own copy or loses access to that book's content.
Where it sits among the billing events
ai-content-monetization sorts the money layer by what event gets billed. Four events were on that list — per visit, per fetch (pay-per-crawl), per use (rsl-standard inference and training terms), per access (wikimedia-enterprise's serviced feed). This is a fifth, and it differs from all of them on who pays:
| Who pays | What is bought | Counterparty needed | |
|---|---|---|---|
| Per fetch | the AI company | one HTTP retrieval | a CDN policy |
| Per use | the AI company | an inference or a training run | a bilateral deal, or unenforced terms |
| Per access | the AI company | a serviced feed with an SLA | a negotiated contract |
| Per copy | the reader | one personal licence to the work | a storefront both sides already use |
Every other row bills the machine. This one bills the human and lets the machine in behind them.
Why that changes the leverage story
The spoke's firmest finding about the money layer is that it pays in proportion to leverage: ~50 bilateral licensing deals since 2023, a crawl toll whose floor rewards whoever is crawled hardest, and one audited case (wikimedia-enterprise, 13 customers) that only exists because the corpus is irreplaceable. Small publishers get nothing from mechanisms that require a negotiation.
Per-copy licensing needs no negotiation over AI rights at all — the transaction is the one the storefront already runs. In principle that reaches any publisher already selling there, which is a wider base than any other mechanism here has managed. Two things temper it, and both matter:
- The storefront is the AI company. Google is the bookseller, the model provider and the party setting eligibility. A mechanism that needs no bilateral deal instead needs the platform's badge on your title, which is a different dependency, not the absence of one.
- Nothing is known about the split. The launch post states no publisher terms beyond the ordinary sale. Whether a book being AI-enabled changes what its publisher earns is unstated, so the "reaches everyone" claim above is a property of the shape, not a measured outcome.
What it applies to
Only works that are already sold as discrete copies — books today, and the post names third-party subscriptions, business research reports and textbooks as next. It offers nothing to a web page, which is the artifact the rest of this spoke's money layer was built for: there is no copy of an article to own, which is why per-fetch and per-use were invented in the first place. Read it as a mechanism for the paid-content half of the problem rather than a general answer.
Related
ai-content-monetization · gemini-notebook-expert-intelligence · pay-per-crawl · rsl-standard · wikimedia-enterprise · crawl-to-refer-ratio · synthesis