Google Is Paying Publishers for AI Training Data. The Debate About Zero-Click Is Already the Wrong Debate.
For two years, the dominant conversation in publishing has been about what AI search is taking away. Zero-click results. Declining referral traffic. AI Overviews that answer the question before the user reaches the source. The frustration is real and the economics are genuinely difficult for publishers who built their business on Google sending them traffic.
But on September 14, 2026, Digiday and Search Engine Roundtable reported something that changes the frame of the conversation entirely. Google has been quietly running a program called the AI Contribution Pilot — a Search Console-based system that pays publishers when their content "significantly contributes" to AI-generated responses across Gemini, AI Overviews, and AI Mode.
The program is small. Dozens of publishers have been invited, not thousands. Payments are based on "value rather than raw usage," meaning Google pays when it judges that content has meaningfully contributed to a response — not simply when content is indexed or cited. The payout interface is basic: a monthly earnings figure in Search Console, with limited visibility into how the calculation is made.
But the mechanism matters more than the current scale. Google is testing a model where the asset being monetised is not a click — it is the inference contribution of the content itself.
What "Inference Contribution" Actually Means
The distinction Google is drawing in the pilot documentation is precise and important. The program explicitly excludes two categories of content use: content that confirms facts after a response is generated, and content that is linked to within AI results. These don't qualify.
What qualifies is content that contributes to the inference phase — the stage where the model is synthesising a response. Content that shapes what gets said, not content that gets cited after the fact.
This is a fundamentally different value proposition from anything the publishing industry has negotiated before. The historical model is: you create content, Google indexes it, Google sends traffic to it, you monetise the traffic. The emerging model is: you create content, Google's AI synthesises responses using it, Google pays you based on how much your content shaped those responses.
The traffic question becomes secondary. The question is whether your content is the kind of content that AI models use to generate answers — original, specific, structured, accurate, authoritative in a domain that the model needs to access to answer real questions.
SEO is not dead — it just stopped being enough. The AI Contribution Pilot makes that argument concrete. You need content that performs in traditional search and content that feeds inference. They are not the same content strategy, but they are closer than the zero-click panic suggests.
What FreeMalta's Numbers Look Like Right Now
I want to be specific here, because abstract arguments about AI search are less useful than actual data from a site that has been building for AI visibility since its founding.
In the last 30 days, FreeMalta generated 100,000+ AI Overview impressions in Google Search Console. On Bing, the same period produced 28,800 total citations across an average of 103 cited pages per day. These are not referral traffic numbers — they are citation numbers. Instances where FreeMalta content contributed to an AI-generated response shown to a user.
I built this deliberately. Getting 5,000 pages indexed in three months required a specific approach to content architecture, internal linking, and technical SEO that prioritised AI systems as a primary audience alongside human readers. The result is a platform that, by the metrics that Google's AI Contribution Pilot appears to value, is already in the right position.
I have not been invited to the pilot. But the logic of what Google is testing maps directly onto what this platform has been building toward.
The OpenAI Approach Versus the Google Approach
The AI content licensing market has two distinct models emerging, and Google's pilot represents the more interesting one for the broader publishing ecosystem.
OpenAI and other AI labs have pursued large lump-sum licensing deals with major publishers — the Associated Press, News Corp, the Atlantic, Dotdash Meredith. These are bilateral negotiations producing fixed payments for broad content access rights. The economics are opaque, the terms are not disclosed, and the deals are primarily available to publishers with enough scale and legal sophistication to negotiate them.
Google's pilot is architecturally different. It is automated, it is integrated into Search Console, it is available to publishers of any size who are invited to participate, and it ties payment directly to measured inference contribution rather than negotiated content access. It is closer to a marketplace than a licensing deal.
One publisher executive quoted by Digiday described it as "an early test of a marketplace for inference data, where publishers learn what's valuable to AI systems as well as to readers." That framing is significant. A marketplace implies discovery — finding out what content AI systems actually use, which is information that most publishers currently have no access to.
The Search Console feedback loop that Google is building — showing publishers which content contributed to AI responses — is potentially as valuable as the payment itself. The ability to see what content drives inference contribution, and therefore what content strategy to pursue, changes how content is made. AI systems are already reshaping which content survives — the pilot gives publishers a tool to understand the mechanism.
Bing's Position in This Landscape
Microsoft's Bing has been the more transparent of the two major search engines about AI citation. Bing's Search Console equivalent provides citation data — exactly the kind of inference contribution visibility that publishers in Google's pilot are asking for. The 28,800 citations FreeMalta received from Bing in the last 30 days are a concrete measure of what inference contribution looks like in practice.
Microsoft has not announced a publisher payment pilot comparable to Google's. But the data infrastructure is there, and the precedent Google is setting will create pressure across the industry to develop similar mechanisms. Once one major platform establishes that inference contribution has measurable monetary value, the others are implicitly acknowledging the same thing by not doing so.
The Cynical Reading and the Optimistic One
Digiday's reporting includes the critical perspective: that Google's pilot "looks less like a meaningful payout and more like a legal fig leaf." The concern is that Google is creating the appearance of publisher compensation without the substance — giving a small amount of money to a small number of publishers to defuse regulatory and legal pressure around AI content use, while continuing to synthesise and display the content of the broader publishing ecosystem without payment.
The concern is legitimate. The pilot is small. The payment amounts are not disclosed. The calculation methodology is opaque. Google's history with publisher economics is not one that encourages optimism about the company's generosity when its commercial interests point the other way.
The optimistic reading, offered by some of the publishers in the pilot, is that the precedent matters more than the current scale. "The fact that we can have an open dialogue, share insights in a meaningful way and have a true partner, I feel like that's so important," one pilot participant told Digiday. They are inside the tent, shaping how the program develops, rather than outside it arguing that the tent shouldn't exist.
Both readings can be simultaneously true. Google is creating a fig leaf and establishing a precedent. The fig leaf is inadequate. The precedent is real.
What Publishers Should Actually Do
The practical implication of the AI Contribution Pilot — regardless of whether any individual publisher is ever invited to join it — is a confirmation of what content strategy should look like in 2026 and beyond.
Content that contributes to inference is original, specific, and structured. It is content that a model uses to answer a question that it cannot answer from general training data — because the question is too recent, too local, too specific to a domain, or requires a point of view that only a particular source can provide. Generic content that rephrases what is already widely known does not contribute to inference. It is already in the model. The model doesn't need to retrieve it.
The publishers who will benefit from the emerging AI content economy are the ones who are building the kind of content that AI systems need to access externally — because it is new, because it is specific, because it reflects genuine domain expertise, or because it covers territory that the model's training data does not adequately address.
The zero-click debate is a debate about what is being lost. The inference value debate is a debate about what is being built. The former is the right question for yesterday. The latter is the right question for now.
FreeMalta is Malta's first AI-native intelligence platform and an Official OpenAI Select Partner. This article reflects the author's personal perspective, informed by FreeMalta's own AI visibility data.
