Google says topical authority is not a thing. Its own documentation is more interesting than that.

What Google says

John Mueller's position has been consistent and dismissive. Search Engine Journal reported him affirming a Reddit deconstruction of the term with a single word, "This", alongside the advice not to worry about it. Google has said there is no official authority score and no topical authority label applied to content.

Worth noting how the denials are phrased. Reporting on Google's statements about site authority records hedges like "from my point of view" and "something like that". That is not the language of a flat denial, and it is the reason the argument never settles.

Meanwhile the term is sold as a deliverable. Topic clusters, pillar pages, and content maps are priced against it.

What the documentation contains

In May 2024 a large volume of internal Google Content Warehouse API documentation became public. Google did not refute its authenticity, and the documents were acknowledged in the course of the United States antitrust proceedings. They have since been republished in browsable form by third parties.

We went to that documentation rather than to the commentary about it. Two fields matter here, and these are their descriptions word for word:

They sit alongside pageEmbedding and siteEmbedding, both described as "Compressed site/page embeddings", and a versionId.

The module holding them is named QualityAuthorityTopicEmbeddingsVersionedItem. Google's public line is that authority is not something it scores. The word sits in the module name.

A correction to the SEO record

Almost every article we read while researching this places these fields in the wrong module. The common claim is that siteFocusScore, siteRadius and site2vecEmbeddingEncoded all live in QualityNsrNsrData, the module associated with Normalized Site Rank.

Checked against the published documentation at the same version, that is wrong on two of the three.

So the three attributes are split across two modules, not grouped in one. site2vecEmbeddingEncoded is in the NSR data, described there as "Encoded site2vec embedding (to be used in superroot) since the full embeddings take too much space". The two topic-focus fields are somewhere else entirely.

This is a small correction and it changes the reasoning. An argument that topical focus feeds Normalized Site Rank is doing work the field locations do not support. We are not claiming the two systems never touch. We are saying nobody has shown that they do, and the usual citation for it does not.

What the fields actually say

Read them plainly and they are narrower than the term built on top of them.

A score for how much a site is focused on one topic is a measure of concentration. A measure of how far a page's embedding deviates from the site's embedding is a measure of drift. Neither is a measure of how good, trusted or expert a site is on that topic.

Focus and authority are different properties. A site that publishes exclusively about one narrow subject scores high on concentration whether the content is excellent or worthless. That distinction is doing a lot of quiet work in how the term gets sold.

Concentration is measurable. Expertise is the thing you actually want. They are not the same variable, and only one of them appears in the documentation.

The patent nobody connects to this

Google holds an active patent on scoring documents by how much new information they add.

US11354342B2, "Contextual estimation of link information gain", assigned to Google LLC, filed 18 October 2018, granted 7 June 2022, and in force until 2039. We checked its status directly rather than relying on a summary.

The patent defines an information gain score as measuring "additional information that is included in the given document beyond information contained in other documents that were already presented to the user". It describes a scale from 0.00, meaning nothing new, to 1.00, meaning the document contains only information the reader has not already seen.

A patent is not proof of deployment. Companies patent things they never ship. But it is a documented statement of intent, it has been maintained through multiple re-grants, and it points in a very specific direction.

Where the strategy breaks

Put the two artifacts together and the standard topical authority playbook is contradicted from both sides at once.

The playbook says: pick a topic, publish a large volume of articles covering every subtopic, interlink them, and authority accrues. Volume is the lever.

• siteRadius penalises drift. Expanding coverage into adjacent subtopics is exactly what moves page embeddings away from the site embedding. The tactic sold to build authority is the tactic that increases radius.
• Information gain penalises redundancy. An article that competently restates what twenty ranking pages already say scores near zero on novelty by the patent's own definition. Publishing more of those adds pages and adds nothing else.

The two pull in opposite directions, which is the useful part. Narrow enough to keep focus high, novel enough to survive an information gain test. A content plan that optimises for volume satisfies neither.

This also explains why the strategy appears to work often enough to keep selling. Sites that publish a lot on one subject do tend to improve. The plausible reason is that some of that content is genuinely new and stays on theme, not that a quantity threshold was crossed.

What we do instead

Three questions before anything gets written, none of which require a content calendar:

• Is this on theme, or is it drift? If a piece would not sit naturally beside the rest of the site, it is widening the radius for whatever traffic it brings. That can still be worth it. It should be a decision rather than an accident.
• What is in this that is not already in the index? A measurement, a test, a document read directly, or two published findings nobody has put together. If the honest answer is nothing, the piece is redundant by the patent's definition and probably by a reader's.
• Would removing it make the site more focused? Pruning is not discussed in topical authority advice, because nobody bills for deletion. On a concentration measure, removing off-theme pages does the same work as adding on-theme ones.

This is the second time we have found the same pattern. Schema markup is sold as an AI visibility lever and the controlled evidence does not support it, while the underlying deliverable remains genuinely useful somewhere else. A name outrunning its evidence is common enough to be worth checking for by default.

This article is a test of its own argument. The information gain in it is the field-location correction and the pairing of the leaked attributes with the patent. If that is wrong, the piece should not rank, and we would rather know.

What this does not prove

Stating this plainly, because the subject attracts overreach.

• The leaked documentation is an API surface description. It shows attributes exist. It does not show they are used in live ranking, how they are weighted, or whether they are still current.
• The documents were published in May 2024. Systems change. Nothing here is dated to today.
• We read them through a third-party republication, not from Google. The field text is quoted as published there.
• An active patent is evidence of intent, not of deployment.
• Google's denials may be narrowly accurate and still leave the practical question open. There may be no attribute called topical authority while site-level topic measurement exists, and both facts can sit together without either side lying.

What we will defend: the field descriptions are quoted correctly, the modules are named correctly, the patent is real and active, and the volume-based playbook is in tension with both. If someone shows the fields have moved, or that focus and NSR are connected after all, we will update this page and say what changed.