The Training Data Dividend — paying creators in the AI era
Executive Briefing 04 · 3 pages · Updated March 2026
The model
The Training Data Dividend is a pay-per-contribution-path model for compensating creators whose work is used in AI training. Rather than a blanket licence fee or a one-off payment, it tracks the contribution of each rights holder's content to the AI operator's outputs and allocates revenue proportionally.
How the waterfall works
Revenue flows through the CIP Rights Registry in a defined waterfall:
- AI operator declares revenue attributable to AI-generated outputs in the relevant reporting period.
- The Training Data Dividend rate (agreed per-operator through the Rights Registry) is applied to that revenue.
- The dividend pool is allocated across contributing rights holders based on their CDR-documented contribution paths.
- Rights holders with collecting society memberships can receive their dividend through existing collecting society channels (PRS, MCPS, DACS, etc.).
- Rights holders without collecting society membership receive directly through the Rights Registry.
Collecting society interaction
The CIP framework does not replace collecting societies — it feeds into them. Where a rights holder has an existing collecting society membership, the Training Data Dividend can route through that society's existing distribution infrastructure. This avoids creating a parallel payment system and respects existing mandates.
For rights holders not represented by a collecting society (common in photography, visual art, and independent publishing), the Rights Registry provides direct distribution.