Google Started Paying for Content That Shapes an AI Answer. Read the Fine Print.

The Google AI contribution pilot is an invitation-only Search Console program that pays a site when its content “contributes significantly” to an AI-generated answer in the Gemini app, AI Overviews or AI Mode. Digiday broke the details on September 14, 2026, and most of the coverage since has treated it as a money story for publishers: who got invited, how small the checks are, whether it is a fair trade. That is a reasonable story. It is not the important one. Buried in the program’s help document is the first time Google has written down, in its own words, what separates content that shapes an AI answer from content that merely gets linked beside one. For anyone doing GEO work, that sentence is the closest thing to an official definition of success that exists, and it comes from the company that runs the three biggest AI answer surfaces in the world.

What is the Google AI contribution pilot?

It is a Search Console feature, offered to a limited set of sites Google chooses, that accrues earnings when a site’s pages are judged to have significantly contributed to a generated response in Gemini, AI Overviews or AI Mode. Google confirmed to Digiday that it is an “early-stage learning pilot to test how best to reward high-quality content, on top of the traffic and tools it already provides.”

The program is not new, just newly understood. Search Engine Roundtable first noticed links to an “AI contribution pilot” in Google’s support documentation on April 13, 2026, and at the time assumed it was a reporting feature along the lines of Bing’s AI performance report. It was not. A publisher who accepts the program terms gets an AI contribution panel in Search Console that shows a monthly earnings figure with some history, a place to manage payment details, and a payments status view that Google warns “may differ from your estimated earnings due to tax deductions or local payment thresholds.” Participation can be cancelled at any time in Search Console settings.

Google’s own framing of the pilot predates the Digiday story. In a June 18, 2026 public policy post, the company said it was “piloting a new way to partner with websites whose content meaningfully contributes to the freshness and factuality of generative AI responses through the process of grounding.” Read that alongside the help document and the shape is clear: Google is testing a payment rail for grounding, the step where a model pulls in live web content to build an answer, and it is testing it outside the news industry, with the kinds of sites that publish product explainers, buying guides and how-to content. In other words, sites that look a lot like yours.

Does Google pay publishers for AI Overviews and AI Mode?

Yes, for the sites inside this pilot, and almost certainly not much. Digiday’s sources describe the payout as a single monthly number with no breakdown of how it was calculated, which one executive summed up as “quite black box.” One source close to the program said early returns were “peanuts” next to advertising revenue. Another publishing executive said the “lowball numbers” offered to their company and several peers did not clear the threshold that would make opting in worthwhile, and that participation could weaken a publisher’s hand in pushing Google for better terms later.

The pilot’s own participants are more sanguine, and their reasons are worth hearing. Two executives inside the program told Digiday that being in Google’s licensing tent, testing direct payments and data-sharing, beats “standing on the sidelines hoping the old referral economics return.” Google is holding weekly calls with some partners. One of them described the pilot less as a licensing deal than as “an early test of a marketplace for inference data, where publishers learn what’s valuable to AI systems as well as to readers,” and said the thing they most wanted was a Search Console-style feedback loop for AI inference.

Whatever the checks turn out to be, the reason the economics matter to a marketing team that will never see a dime is the click data that sits underneath them. Pew Research Center’s analysis of real browsing behavior, published July 22, 2025, found that Google users clicked a traditional result on 8 percent of searches that showed an AI summary, against 15 percent when no summary appeared, and clicked a link inside the summary itself on just 1 percent of visits. That is the world the pilot exists in. When the answer replaces the click, being the source of the answer stops being a nice-to-have and becomes the whole game, and the pilot is the first place Google has put a price on it.

How does Google decide which content shapes an AI answer?

By when the content is used, not whether it is used. That is the part of the help document everyone skipped. Google’s text says the program “focuses on the phase of generation where web content can significantly influence what response is generated,” and then draws the line in one sentence: “Web content may also confirm facts or be linked to within Google’s AI after a response is generated, but those instances don’t qualify for AI contribution.”

Take that apart and you get 3 distinct roles a page can play in an AI answer, only one of which Google is willing to pay for. A page can supply the substance of the response while it is being written. A page can be checked against the finished response to confirm a fact. Or a page can be attached to the finished response as a link. From the outside, the second and third look like success: your URL is in the citation list, your tool records an “AI citation,” your dashboard goes green. Google’s document says the value in those cases was created somewhere else, by someone else’s page, before yours was ever consulted.

This changes the vocabulary. In July we argued that a brand mention and a citation are different outcomes that need to be tracked separately. Google has now added a third rung above both: contribution. Mentioned means the model knows your name. Cited means your page was linked. Contributed means your page is why the answer says what it says. The pilot pays for the third and explicitly not the second, which tells you 2 things at once. Google’s systems can already tell them apart internally, and Google has decided that only the top rung is worth money.

Can you get into the AI contribution pilot?

Almost certainly not, and you should stop trying after one look. There is no application form. Google approaches sites, and Digiday reports it has approached “at least dozens,” a number the publication could not get Google to confirm. The sites that have said yes lean small and mid-sized and span categories well beyond news. Nothing in any of the reporting suggests a way to request an invitation, and the program’s help topic is only visible to accounts already in it. Barry Schwartz, who has covered Google for 2 decades, wrote on September 14 that he still cannot see the document because he is not part of the pilot.

What you can do is use the tool Google did make available to everyone. The AI performance report inside Search Console, which began rolling out in August 2026, shows impressions from Google’s generative AI search features. Digiday quotes one executive who said the report only recently appeared for them, showed “more impressions than they expected,” and closely tracked the traffic they had lost from traditional search and Discover. That is the report to open this week. It will not tell you whether a page shaped an answer or decorated it, but it will tell you which pages Google’s AI surfaces are showing at all, and that is the list to audit against the contribution test below.

One more thing the invite-only status buys you: time. The pilot exists because Google needs to work out how to price a page’s contribution to an answer, “a complicated equation,” as Digiday put it, “no one in the industry has yet figured out.” The methodology will leak, in fragments, through the partners on those weekly calls. Sites that have already sorted their content into shaping pages and decorating pages will be positioned to act on it. Sites still counting citations will be starting from zero.

What does content that contributes to an AI answer look like?

It is the page the model reads first and builds from, which in practice means the page that answers the question in a form the model can lift without rework, on a site the model has reason to trust. Google has not published a checklist, and we are not going to pretend the help document is one. But the generation-phase test has a logic to it, and the logic points at a handful of properties that separate a page that supplies an answer from one that gets attached to it.

The first is that the page contains the answer as a fact, not as a discussion of the fact. A response about heat pumps in cold climates gets built from a page that states the operating temperature threshold and the efficiency figure at that threshold, with the source and date in the same sentence. It gets confirmed against, or linked to, from a page that spends 9 paragraphs on the debate. The second is specificity nobody else has: first-party numbers, test results, a comparison table with real values. A model writing a response has to get the substance from somewhere, and a page that is the only place a particular fact lives is, by definition, a contributor. The third is structure that makes the fact retrievable, the same structure that wins a citation in the first place: a direct answer under a question-shaped heading, a table where the data is tabular, one idea per section. We laid out the anatomy of an AI citation earlier this year, and the contribution test does not replace it. It sits on top of it. You still need to be cited to get read; you now also need to be the page the answer could not have been written without.

The fourth property is the uncomfortable one. Google’s June post says the pilot is about content that contributes to the “freshness and factuality” of grounded responses. Freshness is a property of maintenance, not of publishing. A 2023 guide that has been reworded four times is a citation candidate. A page updated last month with this quarter’s figures is a contribution candidate. If your content operation measures output in new posts per month, the contribution test is going to favor the competitor who measures it in pages kept current.

How do you audit your content for contribution without access to the pilot?

Run the same question through an AI answer surface, read the answer, then ask whether the answer could have been written from your page alone. If yes, your page is a contributor. If the answer says something your page does not say, and your page is merely listed underneath, you are decorating. That is a manual test and it is the only honest one available today, so here is how to run it at scale without fooling yourself.

Start from the AI performance report in Search Console and pull the pages with the most AI-surface impressions. For each one, write down the 2 or 3 questions it is meant to answer, in the words a buyer would use. Ask those questions in AI Mode and in the Gemini app, and save the responses. Now do the part most teams skip: highlight every specific claim in the response, a number, a threshold, a named product, a step, and trace each one back to a source. If the substance traces to your page, mark it “shaped.” If your page is cited but the substance traces to a competitor, a manufacturer or a forum, mark it “attached.” If your page is nowhere, mark it “absent” and note who is there instead.

Three weeks of that produces a list most companies have never seen: the pages that are doing the work, the pages that are getting credit for someone else’s work, and the questions where a competitor owns the answer outright. The fix for an “attached” page is usually not a rewrite. It is adding the one fact the answer needed that your page did not have, in the sentence where a model would look for it. The fix for an “absent” page is deciding whether you can be the best source for that question at all, and if not, spending the effort somewhere you can. That prioritization, more than any single page edit, is what an AI search program should be producing, and it is the part the citation counters cannot do for you.