Google has started paying for content used in AI answers. What that changes for you

For three years the argument about AI answers has run on assertion: publishers said their work was being used without compensation, and platforms said traffic and exposure were the compensation. In the week of 14 September 2026, that argument acquired a number. Google is quietly paying a set of invited sites when their content shapes an AI answer, and showing them a monthly earnings figure in Search Console. Almost everything about how that figure is calculated remains undisclosed.

What is actually confirmed

The separation between what is reported, what Google has confirmed and what nobody knows matters here more than usual, because a pilot with an undisclosed formula is exactly the sort of thing that generates confident advice built on nothing.

Note the shape of that table. The confirmed part is the existence of a payment mechanism. The unconfirmed part is every input a business would need to plan around it. That asymmetry is the story.

Why "contributes significantly" is the interesting phrase

Strip out the money for a moment. The most consequential thing Google has revealed is where it draws the line: reportedly at the generation phase, when web content can shape what a response actually says. A page that merely confirms a fact after the answer is composed, or that appears as a link beside it, reportedly does not qualify.

That is a distinction search professionals have been guessing at for two years, and it is worth taking seriously precisely because Google is attaching money to it. It implies that in Google's own accounting, two pages that both get cited are not equal. One supplied substance. The other was decoration.

What kind of content supplies substance that an answer could not be written without? Not the tenth article restating the same advice in different words — a model composing an answer has that consensus internalised and does not need a source for it. What it cannot generate on its own is specificity: a number with a date attached, a procedure that has actually been performed, an outcome observed first-hand, a constraint that only shows up in practice, a comparison nobody else has made.

The four ways content earns from AI surfaces in 2026

This pilot is one route among four, and confusing them is how businesses end up over-reacting to news that does not apply to them.

For a professional services firm, an e-commerce brand or a local business, the fourth row is the one that pays. Being the source an AI answer draws on shapes what a buyer believes before they ever reach a website, and the fact that Google is now willing to pay for that contribution is confirmation of its value rather than a new way to collect it.

What an AI citation would have to be worth

Nobody outside the pilot can quote a rate, so the useful exercise is the reverse: work out what a payment would need to be for it to matter to you. The arithmetic is simple enough to run on the back of an envelope, and the point is the structure, not the inputs.

Value of a search click   = revenue per session × sessions per click
Break-even AI rate        = value of a search click × clicks displaced
                            ÷ AI answers you appear in

Worked example (illustrative inputs — use your own):
  Revenue per organic session .......... £2.40
  Monthly organic sessions lost to
    answer surfaces on a topic ......... 500
  Monthly revenue at risk .............. £1,200
  AI answers you contribute to ......... 20,000
  Break-even payment per answer ........ £0.06

Six pence per contributed answer, in that illustration, is the point at which payment replaces the lost click. Run it with your own numbers and the conclusion is usually the same: the required per-answer rate is small in absolute terms but large relative to what any advertising-funded system pays for a single impression of a fragment of one source. That is the economic problem this pilot is poking at, and it is why early payments being characterised as small is unsurprising rather than scandalous.

The more useful output of the exercise is the denominator. Most businesses have no idea how many AI answers they contribute to, which means they cannot value the exposure at all. That is the measurement gap worth closing this quarter, whether or not anyone is paying.

What to do now

Three things are worth doing, and two are worth not doing.

1. Establish your AI surface baseline. Search Console's AI-surface reporting, plus referral data segmented by assistant, gives you a rough picture of where you appear and what arrives from it. Capture it monthly now, so that if this pilot ever opens up you can evaluate an offer against something.
2. Audit your content for contribution, not coverage. Go through your top 20 pages and mark which ones contain something an answer engine could not compose without them. Original figures, first-hand procedure, specific dated facts, real comparisons. Pages that fail the test are candidates for a rewrite rather than for another round of date refreshing.
3. Keep your content accessible to the crawlers you want to be cited by. Blocking AI crawlers remains a legitimate business decision, but it is now visibly also a decision about eligibility for whatever compensation mechanisms emerge. Make it deliberately, and revisit it, rather than leaving it to a robots.txt line someone added in 2024.

And the two to avoid. Do not restructure an editorial strategy around an invitation-only pilot with an undisclosed formula; there is no way to optimise for a black box, and the attempt will produce content shaped by guesswork. And do not buy anything that promises enrolment. There is no application process, so nobody can sell you access to one.

What it signals about where this goes

Two readings are available and both are probably partly right. The generous one: Google is testing a mechanism to keep the open web worth publishing on, because an answer engine with no sources left to draw on degrades. The cynical one: a small, undisclosed payment programme is cheap relative to the licensing settlements and regulatory attention that arrive if the supply of willing sources dries up.

Either way the direction is the same, and it is the part worth planning around. The value of content is shifting from being findable to being the thing an answer needed. That change was already visible in how AI systems choose what to cite. What is new this month is that one of those systems has started to put a number on it — small, opaque, invitation-only, and nonetheless a number.

Frequently asked questions

What is Google's AI contribution pilot?

It is an invitation-only Search Console programme, reported from 14 September 2026 and confirmed by Google as an early-stage pilot, that pays participating sites when their content contributes significantly to an answer in the Gemini app, AI Overviews or AI Mode. Participants see a monthly earnings figure in a Search Console panel. Google has not published the payment formula, the eligibility rules, the payout levels or the list of participants.

Can I sign up for the Google AI contribution pilot?

No. There is no application process. Google approached participants directly, reportedly at least dozens of sites, and the programme is not open for self-enrolment. If you have not been contacted, there is nothing to do, and any service offering to enrol you in it is selling something Google does not currently provide.

What does contributing significantly to an AI answer mean?

Per reporting on the pilot, payment accrues when content shapes what an answer says during the generation phase. Content that merely confirms a fact after the answer has been composed, or that appears as a link alongside it, reportedly does not qualify. The practical reading is that supplying substance an answer could not have been written without is what counts, not being one of many pages that restate the same consensus.

How much does Google pay publishers for AI answers?

Google has not disclosed rates, and participants see only a monthly total with no breakdown of how it was calculated. One source close to the programme described early payments as small relative to advertising revenue. Nobody outside the pilot can currently quote a per-answer or per-thousand-answer rate, and any published figure should be treated as an estimate rather than a rate card.

Should I change my content strategy because of this pilot?

Not because of the payments, which are undisclosed, invitation-only and early. The signal is worth acting on: Google is testing payment for content that shapes an answer during generation, which is the strongest available confirmation that originality and substance are valued differently from restatement. That conclusion argues for the content investments that were already defensible, not for a new tactic.

Does the AI contribution pilot replace lost search traffic?

Not at present. Early payments have been characterised as small next to advertising revenue, and the pilot covers a limited set of invited sites. Treat it as a directional experiment in how AI answer surfaces might compensate sources, not as a revenue line that offsets reduced clicks. Plan your content economics on the traffic, demand and brand value you can actually measure.

The takeaway

A monthly number in a Search Console panel, with no explanation of how it was reached, is not a business model. It is a signal, and a fairly loud one: the platform that built its economics on sending clicks is experimenting with paying for substance instead. You cannot join, you cannot optimise for it and you should not plan around it. What you can do is make sure your best pages contain something an answer could not be written without — which was the right editorial instruction before anyone was paying for it, and is now the one with a price attached.