AI search · AEO · GEO
Search is becoming answers, and the answers now pay their sources. Answer engines do not cite opinions. They cite specific, dated, verifiable facts, and the platforms have started paying for the same thing. Data-native content was built for that world before it arrived.
In one paragraph: an answer engine retrieves candidate pages much as a search engine does, then selects the passages it trusts enough to build an answer from and attributes the sources it used. Selection favors passages that are specific, self-contained, current, and backed by a named source. Google's AI contribution pilot pays sites by the same logic: relevance to the query, freshness of the facts, and uniqueness of the content. Content generated from a knowledge graph has those properties by construction. Content written by a language model has none of them, because the model owns no facts, cannot refresh without the data, and invents figures.
What changed
68% of US Google searches ended without a click in January through April 2026. Open-web clicks per thousand searches fell from 374 to 276.
Source: SparkToro, June 2026On queries with an AI Overview, organic click-through runs at 2.36% against 3.82% without one. Being cited inside the answer raises clicks per impression by 120%.
Source: Seer Interactive, April 2026Only 38% of the sources AI Overviews cite rank in the organic top ten, down from 76% nine months earlier; one study of education queries found 17%.
Source: Ahrefs, March 2026; BrightEdge via Search Engine Journal, March 2026Google is piloting payments to sites whose content "contributes significantly" to AI answers, with earnings shown in Search Console. The formula is undisclosed and early payouts are small.
Source: Search Engine Land and 9to5Google, September 17, 2026Brand mentions across the web correlate with AI visibility at 0.66, three times the correlation of backlinks. YouTube mentions are the strongest single signal.
Source: Ahrefs, May and December 2025The same three properties
The pilot's own definition of contribution names relevance, freshness, and uniqueness. The citation research names the same three. Here is how a data-native brand has each one without trying.
| Property | What the platforms pay for | What answer engines cite | What our brands have by construction |
|---|---|---|---|
| Relevance | The closer the content aligns with the specific intent of the query, the higher its contribution. | Query-answer match and the answer near the top are the strongest-evidenced citation factors. | One template per entity, cohort, and place means there is already a page for the specific question. |
| Freshness | Up-to-date facts score higher. | Cited pages are about 26% fresher than the organic top ten; pages not refreshed quarterly are three times as likely to lose citations. | Pages regenerate on every data release with an honest modified date; nothing waits for a writer. |
| Uniqueness | Content not retrieved from other domains scores higher; content that only confirms known facts does not qualify. | Adding statistics raises visibility about 33%, quotations 41%, named sources 28%. | Rankings, percentiles, trends, and value scores computed from primary data exist only where they were calculated. |
What we do for AI search
None of this is a bolt-on. Because pages render from data, a change to the template reaches every page, and because entities are assigned at random, the readout is an experiment.
Prompt sets across the engines with repeated sampling, citation share against named competitors, Search Console AI impressions, Bing AI performance, an index-hygiene audit, and a query-space triage.
Self-contained answer passages near the top, key-facts tables with units and dates, comparisons, methodology and data-vintage statements, consistent entity identifiers, shipped with a randomized holdout.
Prune and consolidate thin pages, protect crawl budget, strengthen internal links, and cover the Bing index that feeds ChatGPT and Copilot.
A monthly original data report and a public changelog of each data release, so newsrooms and answer engines have a reason to return and something new to cite.
An explicit per-bot policy that allows the search and answer crawlers, decides training crawlers deliberately, and enrolls in the programs and marketplaces that pay for content.
Citation share, AI referral conversion, branded search, contribution earnings, and holdout readouts in one report keyed to numbers you own.
How we measure
AI answers are not deterministic, so a single check proves nothing. We run a prompt set of several hundred questions across the engines on a schedule, sample each one repeatedly, and report citation share with confidence intervals against the competitors you name.
Seasonality inflates everything, so a before-and-after comparison proves nothing either. Changes ship to a randomly assigned set of entities while a control group stays untouched. The difference between the two, on citation share, Search Console impressions and clicks, and conversions, is the readout.
We do not grade our own homework. Success metrics are the ones you own: your analytics, your Search Console, your lead system. The methodology and the raw runs are shared, and one independent tool cross-checks our tracker.
What we do not sell
Google does not read it and 97% of published files received zero requests in a May 2026 study of 137,000 domains. We publish one because it costs nothing, and we never bill for it.
A controlled test of 1,885 pages that added JSON-LD found no uplift in AI citations. We emit clean structured data as hygiene, not as a strategy.
Made-up author profiles are rated Lowest in Google's rater guidelines and are banned on LinkedIn. Real named people or an institutional byline; nothing else.
Since May 2026 Google's spam policy explicitly covers attempts to manipulate generative AI responses. We engineer for extraction of true facts, not for tricks.
Every page here carries one structured-data graph describing the organization, the page, and its place in the site, generated from the same facts as the prose. Our robots.txt states an explicit policy for every AI crawler, and llms.txt gives assistants a plain summary of who we are and where to look. Dates on our posts are real publication and update dates. Small things, and the same small things applied to millions of pages are the difference between being read and being cited.
They are overlapping names for the same shift: earning visibility inside AI-generated answers, in Google AI Overviews and AI Mode, ChatGPT, Perplexity, Gemini, Claude, and similar assistants, rather than only in a list of ranked links.
Public reporting in September 2026 describes a Search Console earnings widget that accumulates earnings when content contributes significantly to AI-generated responses, scored on the value of the contribution rather than raw usage. Google has not disclosed the formula, rates, or eligibility rules, and we do not promise earnings from it.
No one can. Answer engines change quickly and disclose little. We build content with every property those systems are known to reward, measure what earns citations against a holdout, and report the readout.
Blocking the search and answer crawlers removes you from the answers; sites that opt out of OpenAI's search crawler, for example, are not shown in ChatGPT search results. Training crawlers are a separate, deliberate decision that can include licensing. We recommend an explicit policy per bot rather than a default.
The baseline takes three weeks and is free for qualified partners. You keep the readout either way.