Topical Authority and Source Selection

Topical authority is practitioner vocabulary, not platform vocabulary — the term entered SEO through agencies and tool vendors, and no search engine issues a score under that name. Coverage is the measurable half of it: the number of distinct questions a source answers inside one subject, each answer on its own document. Reputation is the half the word borrows and does not measure, because links, brand searches and named expertise run through separate systems. A subject's natural questions are the five a buyer asks in sequence: what the thing is, how it works, what it costs, how to verify a provider, and who supplies it locally. Breadth across that sequence is what practitioners point at when a site ranking for one question starts ranking for its neighbours.

The honest caveat: no Google documentation, report or score carries this name.

John Mueller, a Search Advocate at Google, endorsed a Reddit commenter's description of topical authority as a rebrand of relevancy in November 2023, and followed it with the instruction "Don't worry about it."

The 2024 leak of Google's internal Content Warehouse API documentation is the counterweight, and it cuts in both directions. siteFocusScore is defined there as a measure of a site's overall topical coherence. siteRadius measures how far an individual page sits from that site's main topic, computed on site-level vector embeddings. Both attributes live inside the QualityNsrNsrData module, which belongs to a system named Normalized Site Rank. Google confirmed the documents were authentic on 29 May 2024.

Three limits bind every conclusion drawn from that file: the documentation carries no source code, no signal weights, and no marking of which attributes run in production. Subject concentration is measured. The weight it carries is undisclosed.

Answer engines assemble each response from a selected set, never from a single winning page. Surfer SEO, an SEO software vendor, analysed 405,576 Google AI Overviews and measured an average of 5 sources cited per query, with 8 or fewer sources in 90% of cases. Selection is a seat count on that evidence: 5 seats per answer on average, filled from whatever the retrieval step pulled.

Query fan-out is the named technique that widens the pull, and Google Search Central defines it as issuing multiple related searches across subtopics and data sources before the answer is written. Candidate pools multiply as a direct result. A source holding one document on a subject stands in one pool. A source holding five documents — definition, mechanism, measurement, cost, provider selection — stands in five.

Depth raises the quality of one entry. Spread raises the number of entries. Only spread changes how many pools the source appears in at all.

The set covers definitions, per-assistant mechanisms, measurement, cost and provider selection as five linked documents:

• Definitionwhat generative engine optimization is, and how it sits as a subclass of search engine optimization.
• Mechanismhow a source is selected, written once per surface for seven surfaces: Google AI Overviews, ChatGPT, Perplexity, Gemini, Claude, Grok and Meta AI.
• Measurementwhich reports return AI citation data, and how a baseline is recorded.
• Costwhat the work is priced at, and which variables move the price.
• Selectionhow a buyer verifies a provider before signing a retainer.

Contextual links join the five into one set: every document links forward to the next question a reader asks, and the informational documents link into the commercial ones. This site is built to that shape — 56 planned documents split across two sections, 35 informational and 21 commercial, and this page is one of the 35.