About QueryFanout.wiki
QueryFanout.wiki is an independent technical reference and research archive for query fan-out, the retrieval technique Google uses to describe part of how AI Mode in Google Search handles complex questions. The site exists to answer one question precisely: what does query fan-out actually mean, what has Google confirmed, what do the patents describe, what has been tested, and what is still speculation.
It is not a guide to ranking in AI Mode. Several good ones exist and are listed on the Sources page. This site is the reference those guides can cite.
Evidence labels
Every factual claim on this site carries one of five labels. The label is part of the claim. If a sentence has no label, treat it as background prose rather than a finding.
DOCUMENTED Google, or another platform operator, has stated it publicly in writing. The statement is quoted verbatim and linked on the Google page.
PATENT A granted patent or published application describes it. A patent is evidence of claimed engineering, not proof that the production system works that way today.
TESTED We observed it in a logged test that is published in the dataset with a timestamp, so it can be checked and re-run.
INFERRED A reasonable conclusion drawn from documented or tested facts, or a widely repeated practitioner claim that has not been confirmed by the platform. Attributed to whoever made it.
UNKNOWN Nobody outside the platform knows, and we say so. The homepage keeps a running list under What we do not know.
SYNTHETIC Used only in examples and data: a decomposition generated for illustration or planning, never presented as something Google produced.
Editorial rules
- Google's words are quoted, not paraphrased. Where Google has described the mechanism, the exact sentence appears in quotation marks with its source and date.
- Patents are evidence of claims, not of production behaviour. Every patent passage is labelled PATENT and the distinction is repeated wherever it matters.
- Numbers without a public source are labelled as estimates and attributed. Google has not published how many subqueries a fan-out produces. Industry figures such as "8 to 20" are reported as the estimates they are, with the name of whoever published them.
- Observed, reconstructed and synthetic fan-outs are never mixed. An observed fan-out means the subqueries were directly exposed by the platform and captured. A reconstructed fan-out is inferred from retrieval or citation behaviour. A synthetic fan-out is generated for illustration. Each example and each data row says which it is.
- Uncertainty is recorded, not hidden. A claim that cannot be supported is either labelled UNKNOWN or removed.
- Practitioners are credited by name for their claims. Popularity of a phrase is not evidence that its most visible user invented it. Term origins follow the same attribution-confidence approach used by other reference sites in this field.
- No affiliate links, no sponsored placements, no tool recommendations. Tools are described on the Measurement page by what they do and what they cannot do.
- Changes are dated. Each page shows its last-modified date in its structured data, and the dataset is versioned.
How the site is organised
The site is deliberately small. One central concept, a handful of distinct subtopics, each on its own retrievable page, recombined by the homepage into a comprehensive answer. That is a query fan-out architecture applied to a site about query fan-out.
| Page | Job |
|---|---|
| Query fan-out | Canonical definition, mechanism, evidence summary, what is unknown |
| Only what Google has publicly documented, verbatim | |
| Patents | The patent register: subquery generation, filtering, gating, synthesis |
| Examples | Observed, reconstructed and synthetic fan-outs, kept apart |
| SEO | What fan-out changes for content, with practitioner claims labelled |
| Measurement | What can be observed from outside, what can only be inferred |
| Research | Observatory methodology and test log |
| Data | Raw, versioned CSV and JSON |
| Sources | Provenance register for every citation |
Who publishes it
QueryFanout.wiki is an independent reference and is not affiliated with Google LLC or any search platform. It belongs to a family of small reference sites built on the principle that a precise, well-cited page on an exact-match domain is the most useful thing you can give both a reader and a retrieval system. Like the others, it is published under the site's own name rather than a personal byline; the evidence labels and the sources register are meant to carry the weight that a byline would otherwise carry.
Corrections
If a quotation is wrong, a patent number is wrong, a date is wrong, or a claim carries the wrong label, we want to know. Send the URL of the page, the sentence in question, and the source that contradicts it. Corrections that check out are applied within a week and noted in the page's modification date. Disagreements about interpretation are handled by adding the competing view with its own label rather than deleting the original.
Licence and reuse
Original text on this site is available under Creative Commons Attribution 4.0. The dataset on the Data page is released under the same licence. Quoted material from Google, patent documents and named authors remains the property of its owners and is reproduced here under fair use for the purpose of documentation and commentary. You may reproduce the canonical definition freely:
Query fan-out is a retrieval technique in which a search system decomposes a user query into multiple related subqueries, searches for information relevant to those subqueries, and combines the retrieved information into a response.QueryFanout.wiki, canonical definition
- Information retrieval
- AI search
- Google AI Mode