For twenty years the business of a media brand had one shape. Publish a page, rank it, earn a click, sell the attention. Every tool, every team, and every budget line assumed a human reader arriving from a list of ten links.

That reader still exists. But the reader who now decides which sources matter is, increasingly, a machine. And this month the machine started paying.

What was reported this week

On September 17, 2026, several trade outlets reported the details of Google’s “AI contribution pilot,” a program first spotted in Search Console in April. Participants see an earnings widget that, in Google’s words, lets a site “accumulate earnings when your content contributes significantly to AI-generated responses in participating products.” AI Overviews are confirmed; reporting also names AI Mode and Gemini. 9to5Google describes “at least dozens of publishers,” mostly small and mid-sized and “far beyond news publishers.” Search Engine Land notes the widget shows a monthly figure and some history, with no explanation of how it was calculated.

Three details matter more than the headline.

First, the payment is for value, not usage. Search Engine Journal reports that content which only confirms facts does not qualify, and links added after an answer is generated do not qualify either. The line is drawn at the generation step: did your content shape what the answer says?

Second, the money is small so far. One source described early returns as “peanuts” next to advertising. Nobody should re-plan a business around this quarter’s widget.

Third, Google framed the program in June as “a new way to partner with websites whose content meaningfully contributes to the freshness and factuality of generative AI responses.” Freshness and factuality are not marketing words. They are the scoring criteria.

The three words that decide the payout

Participants have seen a plainer description of what is being scored. The unit of payment is a contribution, the share of a generated response that your content accounts for, and it rises with three properties:

• Relevance. How closely the content matches the specific intent of the query.
• Freshness. Whether the facts are current.
• Uniqueness. Whether the content is retrieved from your domain alone. When “multiple domains contribute similar information,” the contribution is split and shrinks.

Earnings are the contribution multiplied by a rate that varies by topic, niche, and location. The rate is undisclosed. The properties are not.

Anyone who has followed the research on AI citations will recognize the list. The original generative-engine-optimization study out of Princeton found that adding statistics raised a page’s visibility in AI answers by about a third and quotations by about 41% (arXiv, 2024). Ahrefs found AI-cited pages were about 26% fresher than the organic top ten (Ahrefs, 2025). A 54-study synthesis this spring ranked factual specificity, self-contained passages, and query-answer match among the best-evidenced factors, and structured data and llms.txt among the weakest (Zyppy, May 2026). What gets cited and what gets paid are the same three things.

Why this matters more than the first checks

The pilot is small. The shift it confirms is not.

In the first four months of 2026, 68% of US Google searches ended without a click, up from 60% two years earlier (SparkToro, June 2026). In education, AI Overviews appeared on 18% of searches in May 2025 and 83% by December (BrightEdge via Search Engine Journal, March 2026). And ranking has quietly stopped implying citation: the share of AI Overview sources that also rank in the organic top ten fell from 76% to 38% between July 2025 and March 2026 as Google adopted query fan-out (Ahrefs, March 2026).

The machines also read far more than they send. Cloudflare’s data through July shows Google sending one visit for every 4.6 pages it crawls, OpenAI one per 217, and Anthropic one per 2,237 (SEOmator, July 2026). That asymmetry is why a market formed: Cloudflare’s pay-per-crawl, Microsoft’s Publisher Content Marketplace, TollBit, ProRata, and Perplexity’s revenue-sharing subscription all exist to price what the crawl-to-click ratio no longer does. Google’s pilot is the largest platform joining a market that was already forming.

So the question for a publisher is not “how much will Google pay me this month.” It is “what kind of publisher gets paid at all, and can I become one?”

Who gets paid, and who does not

Two kinds of publisher fail the three-property test.

The first is the aggregator. If your value was collecting facts that exist elsewhere, an AI answer is a better aggregator than you are, and your contribution is split across every domain that carries the same figures. Google’s 2026 core updates already said this out loud: first-party and originator sites gained, and aggregators, comparison sites, and reference databases lost (Amsive, April 2026; Sistrix, March 2026).

The second is the publisher who fills the gap with content written by a language model. A model owns no unique facts, cannot stay fresh without access to the underlying data, and invents figures when pressed. It fails all three properties by nature, and it carries policy risk on top: since May 2026 Google’s spam policies explicitly cover attempts to manipulate generative AI responses (Search Engine Land).

The publisher who gets paid is the originator. A ranking computed from primary data with a published method. A percentile within a cohort. A five-year trend. A value score. These exist only where they were calculated, they are relevant to the specific question because there is a page for every entity and place, and they are fresh because the pages regenerate when the data does.

That is not a new idea. It is the oldest idea in reference publishing, and it happens to be the one the machines are now willing to pay for.

What to do this quarter

1. Decide your crawler policy on purpose. The search and answer crawlers are how you get into the answers; OpenAI says sites that block its search crawler “will not be shown in ChatGPT search answers” (OpenAI). Training crawlers are a separate, deliberate decision that can include licensing.
2. Enroll where you can. The Google pilot is small and undisclosed, which is exactly when a structurally advantaged supplier should be in the room. Marketplaces are open now.
3. Audit your pages against the three properties. Where is the unique statistic on this page? When was it last true? Is it stated in a self-contained passage a machine can lift, with a source and a date?
4. Measure citations, not just rankings. Search Console now reports AI impressions per page. Run a prompt set across the engines, sample it repeatedly, and track your share against named competitors.
5. Treat contribution earnings as a signal, not a salary. The widget is the first free measurement of which of your content the machine values. Use it to decide what to make more of.

The reader is a machine now. It reads more than it clicks, it pays for what it cannot get elsewhere, and it is very good at telling the difference. Publishers who own their facts have never had a better customer.