The story. Yes — AI training on publisher content without a licensing deal is already showing up in publisher revenue, and the industry’s own executives are now saying the quiet part out loud. In AdExchanger’s Monday roundup, Brian Morrissey of The Rebooting argued that AI “wants information stripped to its raw material” so it can be remixed into something new, and People Inc.’s chief innovation officer Jon Roberts framed the imbalance bluntly: AI needs “chips, you need power, and you need the information to run through them” — and while chip and power supply are well capitalized, the information supply chain is not. USA Today CEO Mike Reed is far enough down that road to say he’d consider blocking Google’s search indexer and AI scraper entirely if Google-driven revenue keeps shrinking. (AdExchanger, Daily News Roundup, 28 September 2026)
01What happened
- A publisher-industry voice named the extraction problem plainly. Morrissey’s “raw material” framing describes what a training run actually does to an article: it discards the reporting, the byline, the context, and keeps only the token patterns.
- A publisher executive quantified the supply-chain imbalance. Roberts’s chips/power/information framing is a reminder that two of AI’s three inputs have trillion-dollar capital behind them; the third — the content itself — mostly does not, unless a publisher negotiates for it.
- A major publisher CEO put a real lever on the table. Reed isn’t just complaining about uncompensated crawling; he’s naming the countermeasure — cutting off the indexer that still sends some referral traffic — as a live option if the trade stops paying off.
- This isn’t an isolated data point. It lands the same week Digiday’s Publishing Summit heard USA Today Media president Kristin Roberts describe Google referral traffic falling from over 70% to roughly 40% of the mix (covered elsewhere in today’s desk) — the traffic side of exactly the leverage problem Reed is threatening to act on.
02What it means inside a GAM network
For an operator running MCM and AdX access on a publisher’s behalf, the Morrissey/Roberts framing matters because it describes why licensing negotiations with AI labs keep stalling on price: a lab buying “raw material” has every incentive to price content as a commodity input, not as the differentiated product a publisher’s ad stack is built to monetize. That’s a category mismatch a publisher negotiating alone will lose. The operator-relevant move is to treat AI-training access the same way a yield desk treats any other demand source: gated, measured, and priced off a floor — not off whatever a scraper is willing to pay when no one is watching the door. Reed’s blocking threat is the tell that publishers with real referral-traffic leverage are starting to price that gate; publishers with less traffic leverage need a different lever, which is where a licensing-aware crawler policy tied into the ad stack’s existing access controls does the work traffic threats alone can’t.
03What publishers should do about it
04The bottom line
Morrissey’s “raw material” line and Roberts’s chips-power-information framing are describing the same mechanism from two different seats: AI training treats content as an undifferentiated input, and it will keep pricing it that way until publishers negotiate like a demand-side gatekeeper instead of a content donor. Reed naming the block-Google option out loud is the first sign that some publishers now have the traffic leverage to make that negotiation real — and the ones who don’t need to build a different kind of leverage before the terms get set without them.