Gannett / USA Today Reformats Content to Win AI Licensing Deals

Gannett is rebuilding USA Today's CMS so machines can read it better: Markdown pages, metadata schemas, machine-readable templates. The goal is more AI licensing deals and a bot allowlist that blocks roughly 99% of crawlers who aren't paying. CEO Mike Reed told the August 6 earnings call to expect more deals this year, with a preference for recurring relationships over one-time checks.

The piece that matters for ad-tech operators is further down the pitch deck. Gannett wants to make branded and sponsored content visible to the large language models behind ChatGPT, and is selling that to advertisers as a new format. The real question the industry should be asking is whether "AI visibility" ever becomes a measurable, bookable ad product with an agreed currency, because that question belongs to the whole sell side, not just Gannett.

The content licensing piece is real but small. High-margin, yes, but the revenue is undisclosed, which usually means modest. The Atlantic and AP signed similar deals without any investor re-rating. Against that sit real engineering costs to re-platform content and ongoing cost to maintain the allowlist. Gannett's leverage is also thin. Local news and sports scores are commodity content, fresh but shallow, exactly what a model can approximate or source cheaply elsewhere. The 99% bot-blocking figure protects future crawls, but the weights already trained on this library exist. That ship has sailed.

The branded-content-in-LLMs angle is more interesting, and further from working. A brand buyer hearing this pitch will immediately ask: how many people saw it, in what context, and can I get a log? Right now the answer is a shrug. Buyers won't move real budget into a channel they can't measure, and the "new format, trust us" pitch has burned them before. What Gannett can sell today is experiment budget, small money, curiosity spend.

The measurement problem is structural. New ad formats don't become bookable until there's a currency buyers trust and a seller can't grade its own homework. The parties who could set that standard fastest are the model providers, and they have no incentive to expose an auditable log of what their answers surfaced. Until an independent measurement vendor gets access to that signal, publishers are selling placements inside answers they don't write, can't see in advance, and can't verify ran.

The reformatting work is sensible hygiene regardless. If AI referral traffic is collapsing and licensing is the hedge, getting the CMS into machine-readable shape is cheap and reversible. Do it either way.

Our call: No standardized, third-party-audited measurement currency for advertising inside LLM answers will be in market by the spring 2027 upfronts. Publisher AI-visibility products will still be sold as bespoke experiments priced by hand. The opposite outcome requires an independent measurement vendor and at least one frontier model to agree on audit access neither has offered. We'll revisit by May 2027. We're wrong if Nielsen, Comscore, DoubleVerify, IAS, or a peer ships an accredited currency for content or ad visibility inside LLM answers before then.

Gannett's USA Today is rebuilding its own website so machines can read it better. Markdown pages, machine-readable templates, metadata schemas, all aimed at winning more AI licensing deals while blocking about 99% of the bots that aren't paying. CEO Mike Reed told the August 6 earnings call to expect more deals this year, and the company wants recurring relationships, not one-off checks. The part every ad-tech operator should watch: Gannett is building a way to make branded and sponsored content visible to the LLMs behind ChatGPT, and pitching that as a new advertiser value proposition.

What's actually being decided: not "should Gannett license content" (that ship sailed) but whether "AI visibility" becomes a real, sellable ad format with a measurement standard behind it. That is the question for the whole sell side, not just Gannett.

Reversibility: Type 2, easily reversed. Reformatting a CMS and managing a bot allowlist are engineering choices you can undo. The strategic bet on a new ad category is stickier, but nobody's locked in yet.

Forcing function: AI referral traffic to publishers is collapsing now, and licensing renewals plus 2027 ad planning are the near-term clocks.

---

The Market Analyst. Two different stories are being stapled together here, and the market only cares about one. Story one: Gannett licenses content to AI companies. That's real but small, and it won't move a stock still defined by print decline and debt service. The Atlantic and AP signed similar deals without any re-rating. Investors won't pay up until there's a disclosed, recurring licensing line with a growth curve, and there isn't one. Story two, the branded-content-in-LLMs angle, is the more investable idea, because it's a new ad format rather than a syndication fee. But new ad formats don't get priced until they have a currency. For a generalist reader: a licensing check is a one-time sale of the library; an ad format is a factory that runs every quarter, and Wall Street pays far more for factories.

The Skeptic. Steelman the bear case, because it's strong. For this to work, AI companies have to keep paying for licensed corpus at scale rather than leaning on data they already scraped, synthetic data, and retrieval. Gannett's leverage is thin. Local news and sports scores are commodity content, fresh but shallow, exactly the category a model can approximate or buy cheaply elsewhere. The 99% bot-blocking figure is discipline theater. The models already trained on this stuff; blocking scrapers now protects future crawls. The weights already exist. And "make branded content visible to LLMs" assumes the LLM will surface it and that an advertiser can verify it did. Neither is true today.

The Operator. Somebody on the ad ops and editorial-tech teams just inherited a mandate they weren't hired for: markdown conversion pipelines, metadata schemas, and a bot allowlist to babysit, all bolted onto a legacy CMS. The 90-day break is the allowlist. Get the logic wrong and you either block a partner you're being paid by or leak content to scrapers you're trying to stop. Neither failure shows up cleanly on a dashboard. Then there's sponsored-content-to-LLM. Brand safety workflows now have to extend into a surface nobody has audited, where you can't control the sentence the model wraps around your client's message. For a generalist: they're promising advertisers placement inside an answer they don't write and can't see in advance.

The Customer / End User (the advertiser). Put yourself in the seat of a brand buyer hearing this pitch. "We'll make your sponsored content visible to ChatGPT." Great. How many people saw it? In what context? Next to what? Can I get a log? Right now the answer is a shrug. Buyers won't move real budget into a channel they can't measure, and they've been burned before by "new format, trust us" pitches with no third-party count. What they might buy today is an experiment line, small money, curiosity budget. That's not nothing, but it's not the display replacement the deck implies. The advertiser demand is real; the proof they need does not exist yet.

The CFO. Look at the actual economics. Licensing revenue is high-margin but lumpy and, so far, undisclosed, which usually means small. Against that sits genuine engineering cost to re-platform content and a permanent cost to maintain the allowlist and the LLM-visibility plumbing. The interesting question is opportunity cost. Every editorial-tech hour spent making pages machine-readable is an hour not spent on the display and commerce business that still pays the bills. This pencils out only if AI visibility becomes a repeatable line item with a price, not if it stays a series of handshake licensing deals.

---

Where the council splits.

The Strategist-vs-Skeptic fight is over Gannett's content itself. Is commodity local news a licensable asset or a giveaway? The Skeptic says shallow content is exactly what models replace cheapest. The bull case says freshness and trust have standing value for real-time answers. Both can't be right about the same library.

The Analyst-vs-Operator split is timing. The Analyst says the branded-content format is the prize; the Operator says you can't sell what you can't measure or keep safe, and the measurement doesn't exist. The prize is real and unreachable at the same time.

The CFO-vs-Customer tension is who pays first. The CFO wants a priced, recurring line before committing spend; the advertiser will only fund a channel once it's measurable. That's a standoff that resolves only when a currency shows up.

---

Synthesis. This hinges on one belief: that "AI visibility" becomes a measurable ad format with an agreed currency. Licensing revenue is a footnote for a debt-heavy publisher and won't re-rate anything. The reformatting work is cheap, sensible hygiene, do it regardless. The real bet, and the one worth watching across the whole sell side, is whether anyone establishes the measurement spec for content and ads inside LLM answers. Whoever writes that standard, publisher, measurement vendor, or the model company itself, owns the category. Gannett is early to the idea and unlikely to own the spec. The leverage in standard-setting belongs to the model providers and the measurement names, and Gannett is a local-news publisher bidding against both.

What to verify before betting on this: is there a disclosed, recurring licensing figure, or just announcements? And has any advertiser paid for LLM-visibility placement against a verifiable count, or is it all experiment budget? Until the second one is yes, this is a positioning story.

---

Prediction: No standardized, third-party-audited measurement currency for advertising or sponsored content inside LLM answers will be in market by the 2027 upfront selling season (spring 2027), and publisher "AI visibility" ad products will still be sold as experiments priced by hand, not as a bookable format.

Confidence: Medium — the demand is real but no measurement primitive exists yet.

Why: The signal in this story is that Gannett is pitching branded-content visibility to LLMs as a new ad product before any way to count it exists, and no player named here (or in the broader cluster) is building the measurement layer. New ad formats don't become bookable until there's a currency buyers trust and a seller doesn't grade its own homework, which historically takes years, not one planning cycle. The people who could set that standard fastest are the model providers, and they have no incentive to expose an auditable log of what their answers surfaced. The opposite outcome, a real audited currency inside 18 months, would require an independent measurement vendor and at least one frontier model to agree on access neither has offered.

Revisit by 2027-05-01: We're right if publisher AI-visibility ad products are still sold as bespoke experiments with no third-party-verified impression or surface count by the spring 2027 upfronts. We're wrong if a named measurement provider (Nielsen, Comscore, DoubleVerify, IAS, VideoAmp, or similar) ships an accredited or widely adopted currency for ad/content visibility inside LLM answers before then.