USA Today Co. and several of the newspapers it owns sued OpenAI in federal court in Manhattan (Southern District of New York) on October 8, seeking more than $250 million in damages. The complaint covers 19 publications and alleges OpenAI copied hundreds of thousands of their articles, without a license, to train ChatGPT. Titles named include The Tennessean, the Detroit Free Press and The Arizona Republic (docket 1:26-cv-08892).
The publishers also argue that ChatGPT can reproduce or summarize those articles in ways that substitute for visits to the originals. The court has not ruled on any claim, and OpenAI did not immediately comment on the filing, according to Reuters. OpenAI has said before that training on publicly available material is fair use. The suit follows OpenAI's news partnerships in Canada and India last month.
A damages claim is only as strong as the record behind it, and most publishers do not yet hold that record themselves.
I learned this building a local media network around federal reporting rules: you win or lose on the record, so the record has to be designed in before anyone asks.
The open question is whether publishers build that record before the next filing, or depend on discovery after it.
Apple has begun blocking companies that track people across the web for ad targeting inside Safari, without warning, using a list it does not publish. Permutive is on the list. Paradium, which owns Men's Journal and Parade, moved its Permutive code through IAB Tech Lab's Trusted Server and is testing whether audience data flows again.
When a vendor stops working in Safari, bids on a publisher's traffic drop, and some publishers already see revenue dips. Trusted Server moves ad processing to publisher-controlled servers, but publishers have resisted it because earlier server-side auctions earned less. Vendors have no clear way to appeal.
The Operator’s Take: Publishers keep audience data only on terms a platform can change overnight. I learned this building scoring models that enterprises used to judge their suppliers: a list people cannot see or appeal is a list they will not accept.
Build · Economic Model Measure what share of your Safari revenue depends on vendors you do not control, and test one Trusted Server route for first-party audience data this quarter.
Swedish start-up Redpine has lined up Sage, the BMJ Group, the Institute of Physics' publishing arm, IGI Global Scientific Publishing, Wanfang Data and RCNi to give AI agents access to more than 20 million peer-reviewed papers. Access runs live through Redpine's servers, so an agent retrieves and cites papers when it answers rather than relying on training data.
Each publisher decides what an agent may access and sets the price per transaction. Redpine says it pays publishers and authors per use, but no detail has been given on how authors are paid or how much.
The Operator’s Take: The part worth copying is that each publisher sets its own permitted uses and price per request; the part still open is the author's share. I learned this opening a whole city's data: access works at scale only when the rules of use are shared, public and enforceable.
Commercial · Economic Model Set permitted uses and a per-request price for your archive now, so the first agent request meets a posted price instead of a negotiation.
At least half a dozen large digital publishers now sell generative engine optimization (GEO) services, Digiday reports. Future has more than 30 clients with renewals, and Ziff Davis adds GEO to branded content deals. Time tested placing an advertiser's message in the markdown files AI agents read; Perplexity then said it would block its agents from Time's site.
Revenue is undisclosed. Chartbeat data in the briefing shows Google Search referrals to publishers down 40.2% year over year, while AI referrals are 0.01% of publisher traffic.
The Operator’s Take: Publishers are selling advertisers a place on the machine's reading list while AI referrals are 0.01% of traffic. The product only holds if the publisher can show what agents actually read.
Build · Detection & Telemetry Record which AI agents fetch your pages and what they receive, so any GEO placement can be priced and proven against actual agent reads.
As search referrals shrink, publishers are asking anonymous visitors to register so they become known users. The Verge plans a hard registration wall for a group of readers later this month. One executive said the rate on registered users "can be as much as eight times" higher with advertisers.
Subscriptions will not cover the gap on their own: one local publisher said its modeling never showed a price point where subscription revenue offset the lost visitors.
The Operator’s Take: A known reader is the one asset an AI answer cannot take, which is why the registered-user rate matters more than the traffic count. The same record also lets a publisher show what its audience is worth when licensing talks begin.
Commercial · Economic Model Price your registered audience separately from anonymous traffic in every advertising and licensing conversation, using a rate card you set.