Google pays 100 publishers for AI answers. Nobody can audit the price.
Google has quietly become a paying customer of the web, and the prices it is paying tell you more about its legal position than its economics: roughly 100 digital publishers are now receiving monthly payments for the content that grounds its AI answers, and for most of them the money is a rounding error.
The Information reported this week that the program — Google calls it the AI contribution pilot, and confirmed its existence to Digiday in September — pays publishers for how much their material contributes to AI Overviews, AI Mode and Gemini. The payment spread is wide enough to read as noise: for several small and midsize sites the totals land under $1,000 across several months, and below 0.1 percent of their ad revenue; at the other end, at least one publisher reportedly clears a million dollars a year. Payments appear as an "AI contribution" panel inside Search Console with a single monthly figure and no breakdown. Participants told both outlets they cannot tell how any of it is calculated, and the amounts move month to month without explanation.
Google's own framing is careful and narrow. In a June policy post, government-affairs vice president Markham Erickson described partnering with websites "whose content meaningfully contributes to the freshness and factuality of generative AI responses through the process of grounding," and called it one of several experiments in value exchange. Set that next to the rest of the same post and the shape of the experiment is obvious: Google already pays flat fees at scale — more than 2,800 News Showcase partners across 33 countries, more than 5,500 European publications through Extended News Previews, 200-plus titles in its News AI pilot. The contribution pilot is the only line where Google pays per use instead of per deal, and it is the only line where the price is unilaterally set, unpaid until after the fact, and unverifiable from outside.
Pricing a document that may be worth nothing
Pay-per-value sounds fairer than pay-per-crawl until you try to compute it. A grounded answer draws on a handful of retrieved documents, and the marginal value of any one of them has no obvious definition: the honest way to measure it is a counterfactual — re-run the same generation with that document removed and see whether the answer changes — at the scale of billions of queries, across a retrieval set that rotates between runs. Nobody in the industry has solved that attribution problem. It is the same measurement gap that makes enterprise AI ROI so hard to pin down that a startup, Ascerta, raised $18 million today just to help companies estimate what their AI spend is producing.
Google's answer to an unsolvable measurement problem is to judge it, and the publisher's answer is to accept the number. Apple has proposed the same usage-linked structure for Siri news, which we examined when the talks surfaced — Apple's pay-per-use news deals could reset AI licensing — but Apple is buying a feature for a product it has not shipped, which gives publishers leverage a live search product does not. Google is paying for content it already crawls, in answers that already run, on a meter it owns and will not explain.
The uncomfortable answer is that the low numbers may be accurate. If a page is one of forty interchangeable documents that could ground a given answer, its marginal contribution really does round to zero, and no pricing scheme will make it non-zero. Which is why the reported earnings cluster around content that is not substitutable — niche verticals with devoted audiences, anime and gaming among them — rather than around the highest-traffic sites. The variable that matters is uniqueness, not volume.
What the payment is actually buying
Run the numbers as a risk calculation and the pilot's design stops looking like stinginess. Against Google sit a September 2025 lawsuit from Rolling Stone's parent Penske Media over traffic and ad revenue lost to AI Overviews, and a formal European Commission antitrust investigation opened in December 2025 into whether Google uses publisher pages and YouTube uploads for AI features without adequate compensation or a real opt-out. Set against that exposure, paying a hundred publishers rounding-error sums is the cheapest possible way to establish that a compensation mechanism exists — cheap enough that the opacity looks less like an oversight than a feature. As publisher coalition founder David Buttle put it to Digiday, per-use payments look like a hedge against a world where Google has to pay, and the infrastructure it is testing now is what it would need if that world arrives.
The structure does a second job, too. By negotiating one publisher at a time, Google turns a collective bargaining problem into a series of bilateral ones, and every publisher who signs becomes the price the next one is offered. One executive in the program told Digiday the concern plainly: participating weakens the argument that publishers are owed better terms, because Google can point to the checks as evidence of compensation. The opt-out Google offers is real but weak leverage — a publisher can now pull its content out of Search's generative grounding, the control we covered when Google hands publishers a button to fight AI search traffic loss, but exercising it costs the audience, which is a far larger number than the payment.
And the payment is arriving against a shrinking baseline. A February 2026 working paper that uses AI Overviews' staggered geographic rollout as a natural experiment found default AIO availability cut English Wikipedia's external search referrals by 5.45 percent versus German and 4.82 percent versus French sibling articles. The pilot money is not replacing that; it is new money, an order of magnitude smaller, attached to a stream that is falling. That is also why the copyright fights keep running in parallel rather than converging — Seattle Times and Newsday sue OpenAI and Microsoft over paywalled news, and every settlement sets a rate the next negotiation works from.
The strongest case for the pilot is the one the program's own participants make: the money is barely the point. What they want is the feedback loop — knowing which of their pages actually influence answers, the way Search Console once taught them what ranked. That is a genuine product, and it is the thing publishers cannot build themselves. It is also the thing Google has the least incentive to make precise.
What to watch
Whether Google publishes an attribution method or grants audit rights to any participant — the moment it does, the pilot stops being a gesture and becomes a market. Whether a large publisher publicly refuses rather than negotiating quietly, which would tell you how much leverage publishers think they still have. Whether the European Commission's investigation forces a payment standard before the pilot's terms harden into precedent. And whether the checks scale past rounding error by the middle of 2027; if they do not, the honest conclusion is that Google ran a two-year experiment in finding out what a citation is worth and the answer came back as not much.
If the value of a page inside an AI answer can't be measured, can it ever be paid fairly — or is the number always going to be Google's to set? Tell us in the comments.
