In late September, Google began paying publishers for content used in its AI products. The reporting came from The Information and was relayed by Search Engine Roundtable on 30 September 2026. Around 100 publishers are in the pilot, covering the Gemini app, AI Overviews and AI Mode.
Most of the coverage focused on the number, which is admittedly striking: several small and midsize sites report receiving one-tenth of one percent of their advertising revenue.
The number is not the interesting part.
The part that should change how you think
Google’s own help documentation, describing what qualifies for payment, says that content which is linked to, or which confirms facts after a response has been generated, does not qualify. Earnings come from content that significantly influences the generation phase.
Read that again with your marketing hat on.
The citation — the little link at the bottom of an AI Overview, the thing every AI visibility tool on the market measures, the thing we ourselves track and report — is explicitly described by Google as not the thing that earns. The thing that earns is upstream and invisible: whether your material actually shaped what the model said.
For the first time, a platform has put a commercial boundary between being referenced and being influential, and has told us they are not the same.
Why this matters even if you will never be in the pilot
If you run a B2B firm in Johannesburg, you are not getting a cheque from Google. The pilot is for publishers, it is small, and by several accounts it is opaque — participants told reporters they do not know how the payment is calculated and that amounts shift month to month without a clear reason.
But the distinction Google has drawn is the useful artefact, because it maps onto a measurement problem we already have.
Every AI visibility product, ours included, measures what it can see. Citations are observable. Mentions are observable. Influence on the generation phase is not. So the entire industry has quietly agreed to treat the observable proxy as the goal, because the alternative is measuring nothing.
Google has now told us, in its own payment terms, that the proxy and the goal are different.
What that does and does not mean
It does not mean citation tracking is worthless. It remains the only observable signal, it correlates with referral traffic you can see in analytics, and a citation genuinely does deliver a visitor. We will keep measuring it and so should you.
It does mean you should hold the number more loosely than the dashboards encourage. A rising citation count tells you that you are increasingly being linked. It does not tell you that you are increasingly shaping the answer — and on Google’s own account, those can move independently.
It also sharpens an argument we have been making for a while. If influence on generation is what matters, then the content that wins is the content a model reaches for while composing, not the content it tacks on afterwards as a source. That is clear, specific, primary material with facts a model cannot get elsewhere: your own data, your own methodology, your own pricing, your own documented process. Not another roundup of someone else’s statistics.
The payout range is its own warning
The reported figures are worth noting for a different reason. Smaller sites earning under $1,000 over several months. One participant at roughly $50,000 to $60,000. One early participant above $1 million a year.
That spread, combined with participants saying they cannot work out the formula, describes a system where the platform sets the terms, the measurement is unilateral, and the supplier has no way to verify what they are owed.
Anyone who spent the last decade on the other side of an ad platform’s reporting will find that shape familiar. It is a reasonable argument for not building a business model on a single channel whose accounting you cannot audit — which is the same argument for owning your email list, your first-party data and your direct relationships, and for treating AI visibility as one input rather than the strategy.
What we would actually do
Keep measuring citations, per assistant, on a fixed prompt set, weekly. It is imperfect and it is what you have.
Stop describing citation counts as influence in client reporting. Describe them as what they are: observable references, a partial view of a larger process.
Prioritise primary material over commentary. If the paid signal is influence-on-generation, the content most likely to carry it is the content that contains something only you have.
Watch whether the pilot expands. If Google extends this beyond publishers, the definition of qualifying content becomes a published, documented ranking signal — and that would be the most concrete guidance anyone has yet given on what AI search actually rewards.
For now it is a hundred publishers, a tenth of a percent, and one very interesting sentence in a help file.