Back in April, some publishers started noticing a new panel in their Search Console that mentioned something called "AI contributions". Google said nothing at the time, and the whole thing was filed away as forum trivia. This week Digiday confirmed what it was. Google is testing paying a group of sites for the content its AI products use to generate answers. Real money, calculated month by month, for contributing to what you see in AI Overviews, in AI Mode and in the Gemini app.
We have spent three years debating whether AI companies should pay for the content their models consume. Google just answered yes, even if it did so quietly and with a guest list. Let's go through it piece by piece, because the details matter more than the headline.
What is the AI Contribution Pilot?
The program is called the AI Contribution Pilot and it works by invitation. According to the reporting, Google has approached at least several dozen publishers, with an interesting skew toward small and mid-sized sites, and not only news sites. An invited publisher accepts terms, sets up a bank account, and from then on sees a monthly earnings panel in Search Console, historical data included. They can opt out anytime from the settings.
Google's official description is the usual one for these cases, cautious to the point of boredom. They call it "an early-stage learning pilot to test how best to reward high-quality content", and say it targets sites whose content adds "freshness and factuality" to their AI answers. Write those two words down, freshness and factuality, because we are coming back to them at the end.
When does Google pay, and when does it not?
This is the most interesting part of the pilot, and the least discussed. Google is not simply paying you for showing up as a citation in an AI answer. The payment kicks in when your content "contributes significantly" to the generation of that answer, meaning the model used your page as an input while it was putting the text together.
The distinction has concrete consequences. If your content only confirms facts after the answer has already been generated, you don't qualify. If the AI links to you as a source at the end but didn't use you to write the response, you don't qualify either. In practice, Google is paying for being an ingredient of the answer, which is quite different from being the link that shows up at the end.
And how much does it pay? Nobody really knows. One executive participating in the pilot described the calculation as "quite black box", with no visibility into how the amounts are determined. Another participant summed it up with a candor I appreciate: "Do I wish they were more transparent? Definitely". Add to that Google's warning that the actual payment may differ from the reported earnings due to taxes or minimum thresholds depending on the country. In other words, it is a real pilot, with all the rough edges of a pilot.
Why is Google doing this now?
Because the pressure reached a point where ignoring it costs more than paying.
The context is familiar if you follow this blog. A recent analysis of traffic to the 50 largest news sites in the United States showed year-over-year declines at 41 of them, with Newsweek, Daily Mail and Business Insider among the hardest hit, and the most cited explanation is AI answers resolving the query without generating the click. Publishers are suing Google over exactly this. Cloudflare launched its pay-per-crawl system. OpenAI, Amazon and others have spent months signing licensing deals with media companies. The entire market is moving toward putting a price on the content that feeds the models.
And there is a less visible but more structural reason. Google's AI answers need fresh, verifiable content to work. If the sites producing that content lose their business model, or decide to block Google's crawlers, the quality of AI Overviews and Gemini degrades. Paying a fraction to the suppliers of the input is cheaper than running out of input. Seen that way, the pilot has more to do with securing a supply chain than with goodwill.
What does this mean for SEO and GEO?
For me, the money is the flashiest part of this, but what the money reveals seems even more important.
First, Google can measure how much each document contributes to each generated answer. We already suspected this since the AI performance report in Search Console appeared, showing impressions and clicks from AI experiences. But this pilot confirms there is an internal contribution metric reliable enough to move money to third parties. When a company turns a metric into payments, it is because it trusts that metric.
Second, a new unit of value shows up in the ecosystem. For twenty years, the value of a page was measured in the clicks it receives. Now there is, even if only in a pilot, a value for what your page contributes to an answer that may never send you traffic. Those are different things, and you optimize for them differently. A page can be mediocre at generating clicks and extremely valuable as AI input, because it holds the exact, updated, verifiable piece of data the model needs.
Third, this puts a floor under something we in GEO had been arguing with more logic than evidence: that being source material for AI answers is an asset. Now there is one player, the largest of all, willing to pay for that asset. You can argue about the amounts and the opacity, but you can no longer argue about whether it exists.
What can you do if you are not invited?
You cannot apply to the pilot, it is invitation only. But the criteria Google says it is rewarding, freshness and factuality, are within your reach starting tomorrow. Here is what I would do this week.
First, open the AI performance report in your Search Console and check which of your pages are already showing up in AI experiences. That is your baseline, and it is free.
Second, take inventory of your citable content. Pages with original data, dated figures, clean definitions, explained methodologies. That is the kind of content a model uses while generating an answer. An opinion piece without a single verifiable fact can be great for your brand, but it makes poor input for an AI assembling a factual response.
Third, publish something only you can publish. If you run an ecommerce site, aggregated data from your own operation. If you are a consultant, an industry benchmark with the methodology in plain sight. As an illustrative example, a report along the lines of "we measured the load time of 80 retail sites in Chile" has a far better chance of becoming answer input than the umpteenth article on "what is load time".
Fourth, review your stance on AI crawlers with data rather than by reflex. Blocking them protects your content, but it also takes you off the field where this pilot suggests value is going to be distributed. There is no single right answer, it depends on your business model, but make it an informed decision and not an inherited configuration.
If the pilot scales, the sites that have spent time building content with fresh, verifiable data will arrive with an advantage to a licensing market that is only now taking shape. And if the pilot dies, as so many Google pilots do, that same content is what gets you cited and recommended by AI today. It is one of the few bets in this field that pays off in both scenarios, so this week's to-do list looks fairly clear.