The one decision everything else follows from
Every SEO tool has to answer one question about queries: where do they come from?
There are two honest answers. You can generate them, by modelling a topic and deriving the queries a page should cover. Or you can observe them, by reading what Google Search Console reports the page already receives.
We observe. Every query in WebmasterNinja comes from the Search Console API, authorised by your own read-only token. No query is invented, and no keyword appears that Google did not report.
This is not a claim that observing is smarter. It is a trade, and the costs are real. They are listed further down.
What observing gets you
A page's own query footprint is evidence, not opinion. When we say a term matters for a page, the claim is reconstructible: Google sent this page 340 impressions for that term last month. You can check it in your own Search Console within about thirty seconds.
That property is what lets the tool refuse to guess. If the data does not support a suggestion, the panel stays empty. An empty panel is not a bug in a tool built this way. It is the tool working.
It also means the two features people find most useful are cheap for us and awkward for everyone else. Demand match compares the vocabulary Google already associates with your page against the words actually on it. Near-duplicate detection finds two of your pages competing for the same query footprint. Both read the same measured source.
Where the patents actually point
The idea that covering a topic's related terms helps is not folklore, but the usual justification for it is. "LSI keywords" are not a Google mechanism, and no amount of repetition makes them one.
The grounded version sits in US7536408B2, Anna Patterson's phrase-based indexing patent. A document is scored partly by how many of a topic's related phrases it contains, where a phrase counts as related when its actual co-occurrence with the topic exceeds its expected rate. That is information gain, and it is a described retrieval mechanism rather than a rumour.
Near-duplicate detection has its own patent, US8108412B2, which fingerprints a document by its high-value related phrases. We proxy that fingerprint with the page's real query footprint, because the query footprint is the part we can actually observe.
The distinction we care about: the patent describes phrases a document contains. It does not license a tool to invent the phrases a document should contain.
Where we differ from Koray's framework
The most developed alternative is the semantic SEO framework taught by Koray Tugberk Gubur, widely credited with formalising topical authority as a working method. It is a serious system, and this is a comparison rather than a rebuttal.
In that framework you begin before the data exists. You define the source context, the purpose of the site and how it earns. You pick a central entity and a central search intent. From those you derive a topical map, a structured network of core and outer sections, and a query network spanning query aspect, definition and theme. Broad representative queries stand in for the narrower represented queries beneath them. Then you write against the map.
Three differences follow, and only the third is a disagreement.
Unit of analysis. That framework operates on a site and a content network. We operate on the single page you are looking at, in the browser, while you look at it. Neither unit is correct in general. They answer different questions.
Direction. It is prescriptive and runs before publication. We are descriptive and run after. A topical map tells you what to build. Our query footprint tells you what happened when you built it. A team that only uses our tool will improve pages that already receive traffic and will get no help planning pages that do not exist yet.
What counts as measurable. This is the real disagreement, and it is narrow. There is no topical authority score here, and none is planned.
Why there is no authority score
The reasoning, which you are free to weigh differently:
- Google's published "Topic Authority" (Search Central, May 2023) is scoped to news, and its stated signals are largely off-page: notability, citations of original reporting, source reputation. It is not a general per-topic score exposed to site owners.
- The topical-authority patent most often cited in SEO discussion, US11874882B2, is Microsoft's, not Google's. Google's site-level quality patents that we could find, including US9767157B2 and US8682892B1, describe topic-agnostic quality, not authority per topic.
- Topic clusters as a pillar-and-spoke discipline trace to HubSpot's 2015 to 2017 work, not to Google documentation.
None of that shows the practice fails. Sites built this way rank, and we are not disputing the outcomes. It shows we cannot compute the number honestly from data we can see, so we do not print one. The mechanism underneath clustering that Google does document is plainer and duller: internal linking, descriptive anchors, no orphan pages. We report that instead, as evidence rather than a score.
This is also why our coverage tab carries no numeric ring. It shows which subtopics your site covers and where the gaps are, drawn from your own query data, and it stops there.
A worked example of the bar
We recently tried to label individual queries with intent. The plan was a classifier that reads a query string and marks whether it wants fresh content, a deeper answer, or reflects surging demand.
The freshness and depth halves were regex classifiers. On a sample of fifty real Indonesian queries, the depth label fired on forty-two. We reviewed those by hand and the precision landed around 0.81, against a bar of 0.80 set before the test.
Then two independent reviews found the number was not trustworthy. We had judged by term class rather than query by query, so the effective sample was four decisions, not forty-two. The confidence interval at that sample size ran from roughly 0.66 to 0.90, which cannot distinguish passing from failing. And the corpus was built to study grammar, so it was unusually friendly to the classifier.
The label did not ship. What did survive is the part that needs no interpretation: a surge label computed from impressions, where the tooltip states the arithmetic and you can check it.
We mention this because it is the whole method in one story. Anyone can build the classifier. The bar is whether you turn it off when your own evidence does not clear the line you set.
Which to use
If you are planning a site that does not exist yet, or entering a topic where you have no history, a topical map gives you something to build against and we give you nothing. Use the map.
If you have pages that already receive impressions and you want to know which ones are underperforming their demand, which of your pages are competing with each other, and what a specific page fails to answer, that is the question we are built for.
Most people doing this work need both, at different times.