Google Now Pays for AI Citations, and the Search Console API Cannot Measure Them
Table of Contents
In one line: Google has started paying for content cited in AI answers, but the money is small, entry is invitation-only, the formula is undisclosed, and the Search Console API currently returns no data on the thing being paid for. The test and the workaround are below.
On September 14, 2026, Search Engine Roundtable reported Google's AI contribution pilot, with Digiday and Search Engine Land covering it the same day. It is worth a look for small site owners specifically, because small sites are who it targets.
What the programme actually does
When your content "significantly contributes" to an AI answer in AI Overviews, AI Mode or Gemini, Google pays you. It runs through Search Console, where joining adds a widget showing a monthly earnings number and not much else.
Google's statement to Search Engine Land: this 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."
Key characteristics:
- Value, not volume. Google pays when it judges a piece of content meaningfully contributed to an answer, not per crawl or per citation.
- Undisclosed formula. How payment is calculated is not explained. Search Engine Land describes it as quite black box.
- Invitation only. At least dozens approached, with no published application route.
- Not just news. It has widened beyond news publishers and appeals more to small and mid-sized sites than to large ones.
What publishers say about it
Not much that is flattering. Publishers quoted by Digiday described the payments as peanuts, too low to clear the bar for participating.
The structural criticism matters more. David Buttle, founder of Spur, called it a kind of hedge, a test run for a future in which licensing might be mandatory. Other publishing executives described it as the illusion of contribution, serving a regulatory purpose while dividing publishers through selective enrollment.
Those criticisms point at the same thing: the function of the payment may not be compensation, so much as establishing the fact that payment happened.
What I tested: the Search Console API cannot measure it
Everything above is other people's reporting. This part I ran myself.
Scope first: I tested the Search Console API, not the web interface. Reporting notes that an AI performance report landed in the interface in August, so the interface does have something. What I tested is whether you can pull this out programmatically and check it yourself. You cannot.
If Google starts paying based on a content's contribution to AI answers, the obvious next question is whether you can see how often your content gets cited. I tested the Search Console API against our own property.
You cannot.
Range 2026-08-17 to 2026-09-13
searchType:
web clicks 1432 impressions 63939
image clicks 0 impressions 17
news clicks 0 impressions 2
discover clicks 0 impressions 0
aiMode HTTP 400 Invalid value at 'type'
aiOverview HTTP 400 Invalid value at 'type'
searchAppearance:
TRANSLATED_RESULT impressions 29
FORUMS impressions 948
Three things:
- Passing
aiModeoraiOverviewastypereturns HTTP 400 saying the value is invalid. It does not return empty data; it does not recognise the dimension. - The
searchAppearancedimension contains only translated results and forums. No AI entries. - Positive control: the same script, same range,
webreturns 1,432 clicks and 63,939 impressions. This step matters. It proves the query works, which is what makes the absence above meaningful. Ifwebhad come back empty too, that would mean I queried wrong, not that the data is missing.
That control was deliberate. When you find nothing, "there is nothing" and "I asked wrong" look identical, and an empty result means nothing until you have shown the query fires.
To rerun it: post your property and date range to searchAnalytics/query, swap type for the value you want to test, and always include one type: 'web' call as a control.
So the situation is
Google has started paying for something you cannot compute for yourself.
This is probably not a conspiracy, just product sequencing: the interface shipped first and the API has not caught up. But for a site owner the effect is the same: you receive a number you have no way to check.
This is the same class of problem I have been cleaning up on our own site. Last week I had a set of platform numbers on a public page labelled "live" that were actually hardcoded constants; a day later the "live" account count was off by ten. The fix was to make the section genuinely read the public endpoint and hide the whole line when the endpoint cannot be reached. Better one line fewer than a number nobody can verify.
Apply the same standard here: an earnings figure you cannot recompute should not become an operating metric.
What to look at instead
Look at what you can actually measure: referral traffic from AI assistants in your own analytics.
These hosts show up in your source reports: chatgpt.com, gemini.google.com, claude.ai, perplexity.ai, copilot.com, copilot.microsoft.com, kagi.com and others. The list keeps growing, so match on keywords rather than enumerating.
Our three sites over the last 28 days. Monthly sessions across them run from roughly one thousand to four thousand:
| Site | Organic search share | AI assistant referral share |
|---|---|---|
| ultralab.tw (mostly a technical blog) | 54.0% | 0.69% |
| ultra-advisor.tw (tool-led site) | 72.6% | 0.94% |
| mindthread.tw (product site) | 31.3% | 1.76% |
All three sit around one percent, and the more product-led the site, the lower its organic share and the higher its AI referral share. The honest limitation: this only captures cases where an AI linked out and a person clicked. It misses every case where an AI read your content, answered fully, and nobody clicked. That second group is almost certainly larger.
But it has one decisive advantage: it is real, it reruns, and it compares to last month. Until Search Console produces AI data, a limited true metric beats a complete-looking one you cannot check.
Three things a small site should do now
One, do not reshape your content strategy for this money. The amounts are a rounding error, and you probably have not been invited anyway.
Two, add AI assistant referrals to the report you look at every month. Add it now, because what you need is a baseline. When AI traffic does arrive, without a baseline you cannot say how much it grew.
Three, keep writing content in a shape that can be cited. Question-and-answer structure, specific numbers, named sources, FAQ structured data, an llms.txt. These are worth doing whether or not Google pays, because they determine whether you appear in the answer, and appearing in the answer is the actual prize. The payment is a side effect.
If you want to see how your own site looks to AI engines right now, our free check takes a URL and returns the gaps across SEO, AEO and AAO: ultralab.tw/probe.
One thing to remember
The number to take from this programme is not the payout. It is the precedent: being cited by an AI is now a thing with a price.
The price is low today, the formula is opaque, and the measurement is missing. All three will change. Whoever starts measuring what is measurable now will be standing in a better place when the rules get clear.
This is a lab weeknote, not a product promise.
Sources
- Search Engine Roundtable, Barry Schwartz, 2026-09-14
- Digiday, Google rolls out pay-per-value AI licensing program to publishers
- Search Engine Land, Google tests paying publishers for using its content in AI Mode, AI Overviews and Gemini
- Search Console API test run against this site's property, 2026-08-17 to 2026-09-13
- GA4 data for three properties, last 28 days, pulled 2026-09-16
FAQ
What is Google's AI contribution pilot?
An early-stage Google pilot that pays site owners when their content significantly contributes to AI-generated answers in AI Overviews, AI Mode and Gemini. It runs through Search Console, where joining adds a widget showing a monthly earnings figure. Payment is based on value rather than raw usage, and the calculation is not disclosed. It was spotted in April 2026 and confirmed by Google in September 2026.
Who can join, and how do you apply?
It is invitation-only with no published application route. Digiday and Search Engine Land report that at least dozens of publishers have been approached, that it has widened well beyond news sites, and that it appeals more to small and mid-sized publishers than to large ones. Without an invitation there is currently no way in.
How much does it pay, and is it worth changing strategy for?
No. Publishers quoted by Digiday described the amounts as peanuts, too low to justify participating. For a site with a few thousand organic clicks a month, any payment at that scale is a rounding error. Its value is as a signal, not as revenue.
Can you see how often AI cites your content in Search Console?
Not through the API. Testing our own property, passing aiMode or aiOverview as the searchAnalytics type returns HTTP 400 invalid value, and the searchAppearance dimension contains no AI-related entries. The web control in the same query returned 1,432 clicks, so the query works and the absence is real. Any claim about AI citation counts today cannot be backed by Search Console data.
What should you track instead?
Referral traffic from AI assistants in your analytics: chatgpt.com, gemini.google.com, claude.ai, perplexity.ai, copilot.com and similar. Across our three sites over 28 days it runs between 0.69% and 1.76%. It only captures cases where an AI linked out and someone clicked, so it undercounts, but it is real, repeatable and comparable month to month.