I went looking for a straight answer on topical authority and found two from Google that point in opposite directions. One says the thing does not exist. The other is a leaked schema with three fields that describe it precisely. I spent a day deciding which to believe before working out that the question was wrong.
This is the first of a series where I try to build topical authority on my own sites and publish the numbers either way. Part two is the baseline audit.
What is topical authority?
Topical authority is how completely and consistently one site covers one subject, measured across pages rather than inside a single page. A site with one post about email deliverability has a page. A site with fifteen linked posts covering deliverability, warm-up, segmentation, bounce handling, and compliance has topical authority on email. The unit of measurement is the site, not the article.
That distinction is the whole thing. I had been thinking about SEO one page at a time. Write the page, pick the keyword, move on. Topical authority says the page is not the unit. The site is. Which means a good post about something unrelated to everything else you publish can actively cost you, and that was not how I had been thinking at all.
Why does Google say it does not exist?
Because Google ships no metric called topical authority, and its own SEO Starter Guide is blunt about several other things SEOs treat as levers that it says are not. In a Reddit thread, a commenter argued the concept was not new and was really just relevance under a different name. John Mueller, then a Google Search Advocate, agreed with that reading. He was not announcing a ranking system. He was agreeing that the vocabulary was recycled.
My first reaction was that this settles it and the term is marketing. It is a reasonable read, and I want to be careful not to overstate what happened: this was one Google employee agreeing with a stranger on a forum, not a documented position. Then I went and looked at the leak, which complicated it.
What was in the 2024 Content Warehouse leak?
The 2024 Content Warehouse API leak — first dissected in detail by Mike King at iPullRank — exposed 14,014 attributes across 2,596 modules, and among them sit site-level fields that, per the documentation, describe what SEOs mean by topical authority: siteFocusScore, how concentrated a site is on one subject; siteRadius, how far a site’s pages deviate on average from that center; and siteEmbedding, a compressed vector of what the site is about. Presence in a warehouse schema is not proof of use in ranking, and these documents describe storage rather than a ranking formula.
I put that caveat inside the paragraph rather than after it on purpose. If an answer engine lifts one block from this page it will lift that one, and a hedge sitting three paragraphs later does not travel with the quote.
What is still notable is the shape of the fields. Focus, radius and embedding describe a site the way you would describe a physical object, with a center of mass and a spread around it. Whatever Google calls it, something in there is modeling what a site is mostly about.
Is topical authority real or not?
Both answers hold. There is no score called topical authority, exactly as Mueller said. There are site-level topic representations that produce the behavior the term was coined to describe. The disagreement is about vocabulary, not about whether concentration matters.
Once I framed it that way the practical answer stopped depending on who was right. Whether the mechanism is called relevance or focus, the same thing follows: publishing across scattered subjects works against you, and publishing inside one compounds. I do not need Google to confirm a metric name before acting on that.
What does this change about what I publish?
It changes the question asked before writing. I used to ask whether a post was good. Now I ask whether it sits inside the shape of what this site is already about, because a good post outside that shape pulls the site’s center away from where I want it.
Reading my own Notebook with that lens was uncomfortable. There are posts on agent frameworks, on eval scores, on picking the first thing to automate, and on why an integration broke in week two. Individually I stand behind them. As a set they are held together by me having written them rather than by a subject.
Which means this post has a problem. A series about SEO, sitting on a site about AI automation for small businesses, is the exact off-shape publishing I just described. I am doing it anyway, because the alternative is writing about topical authority without ever testing it, but it belongs on the record that I am widening the radius while complaining about the radius.
Does any of this survive AI search?
It matters more, because answer engines have to pick which sources to trust for a subject and have less room to hedge than a page of ten blue links. A results page shows ten sites and lets you choose. An AI answer names two or three, or none.
Ahrefs’ study of 75,000 brands found branded web mentions correlate with AI Overview visibility at a Spearman coefficient of 0.664, against 0.218 for backlinks. That is a real published study with a stated method, and I said in an earlier draft that I could not trace it, which was me not looking hard enough. Two caveats that matter more than the numbers: correlation is not causation, and both mentions and links scale with how big a brand already is, so some of that gap is just size showing up twice.
What should you do with this?
Before your next post, write down the one subject your site should own, then check whether the post you are about to write sits inside it. If it does not, you are not being punished for writing it, but you should know you are paying for it.
I am going to run that test on myself in public and publish what it costs. If you would rather have that audit run on your own site than read about mine, that is the AI-visibility work I sell. Part two is where I count what I am starting with.