SEO · 10 min read
Topical authority is a working model, not a score you can raise
Nobody outside Google can point to a topical authority setting, and the term is used to justify a great deal of publishing that changes nothing. What the model genuinely predicts, which parts of it are supported by documented behaviour, and how to act on it without producing forty articles nobody reads.
Written by Zubair Afzal, FounderUpdated
Nobody outside Google can point to a topical authority setting. What they can point to is a pattern, and the pattern is worth acting on even though the mechanism is not published.
That distinction is the whole article. If you treat topical authority as a confirmed ranking factor, you get a plan built on a claim nobody can support, and you end up publishing to fill a map. If you treat it as a model that makes testable predictions about coverage, connection and expertise, you get a plan that is defensible whether or not the underlying mechanism is what anybody thinks it is.
Google’s published guidance never uses the phrase. It does ask whether a site has a primary purpose or focus, whether content is produced by people with real knowledge of the subject, and whether a reader would feel they had to search again after reading. Those are three different questions, and a site that answers all three well would exhibit exactly the behaviour practitioners describe as topical authority.
The model
What the idea predicts, stated so it can be wrong
A model is only useful if it says something that could turn out to be false. Stated carefully, this one makes three predictions. First, that a set of pages which between them answer every reasonable question about a subject will compete better within that subject than the same effort spread across unrelated subjects. Second, that connection matters — pages that reference each other in the order a reader would need them outperform the same pages published in isolation. Third, that a site with demonstrable knowledge of a subject will find new pages in that subject easier to establish than a site starting cold.
The first two are things you can arrange deliberately. They are also, conveniently, things that make the site better for the person reading it, which means the work is worth doing on its own terms even if the search benefit turns out to be smaller than claimed. That is the strongest argument for acting on this model: the downside case is a genuinely better library.
The third prediction is the one people most want and control least. Whether a site is treated as knowledgeable about a subject depends partly on things you do not decide — who references you, what is said about you elsewhere, whether the people writing are identifiable and actually know the field. A plan built only on publishing has no mechanism for that half at all.
Where it goes wrong
Four ways this idea produces bad work
Almost every disappointing outcome we see traces back to one of these, and all four are avoidable before a single page is commissioned.
- The map came from a keyword export, so nobody can say who each page is for.
- A generated map contains nodes that exist because a phrase exists. Publishing them produces pages nobody searched for in that shape, which then have to be maintained forever. The fix is to write the questions before the map: if you cannot name who asks a question and what they do next, that node is not a page.
- The subject boundary was never drawn, so the plan keeps growing.
- Without a stated boundary, every adjacent subject looks like part of the cluster, and a plan for one subject becomes a plan for the whole industry. Drawing the boundary is the hardest decision in this work and the one most often skipped. It should be written as a sentence: what this site is the best place on the internet to read about, and what it explicitly is not.
- Pages were published on the same subject but never connected to each other.
- Twenty articles that do not link to each other are twenty isolated pages, not a body of work. Connection is the cheap part and it is skipped because it happens after publication, when attention has moved on. Each page should link to the pages a reader would need before it and after it, using anchors that describe the destination rather than repeating the target phrase.
- Volume was used as a substitute for having something to say.
- Content assembled from the first page of results can only summarise what already ranks. This is the ceiling that publishing harder does not raise, and it is why a competitor with fewer, better-informed pages frequently wins. If nobody in the business can add something a researcher could not find elsewhere, the honest response is to change the subject rather than the volume.
A method
Five steps that produce fewer pages, not more
The discipline here is subtractive. Most of the value comes from the things this process removes from the plan.
Write the boundary as one sentence
What is this site trying to be the best place to read about, and what is explicitly outside that? The sentence has to exclude things, or it is not a boundary. A boundary that excludes nothing produces a plan that never finishes and a site with no identifiable subject.
You get: One sentence, agreed by whoever signs off the budget
List the questions, not the keywords
Every question a buyer, a beginner and a sceptic would ask inside that boundary, written as questions in their words. Draw them from sales calls, support tickets and the people in your business who answer them daily, then check the list against a keyword export to catch phrasings you missed rather than to generate the list in the first place.
You get: A question list with the source of each question noted
Collapse the list until each page is unambiguous
Merge every question that would be satisfied by the same page. Then remove any question you cannot answer better than what currently ranks for it, and any question where nobody in the business has first-hand knowledge. What survives is the plan, and it will be considerably shorter than the list you started with.
You get: A page plan, plus a removed list with reasons
Decide the order of reading, then link in that order
Arrange the surviving pages as a reader would encounter them: what has to be understood first, what follows, what is a dead end that only some readers need. Then write the internal links to match that order, with anchors that describe where the reader is going. This is what turns a set of pages into a body of work, and it takes an afternoon.
You get: A reading order and an internal link plan
Publish, then maintain rather than extend
When the set is complete, resist adding to it. The next useful action is usually keeping it correct: updating what changed, merging pages that turned out to overlap, and removing anything nobody reads. A complete and current set outperforms an ever-growing one, and it is the state most content programmes never reach because there is always another node on the map.
You get: A review date per page and a named owner for the set
Inputs
Four inputs, and how much of each you control
These are usually presented as one programme. They are not: they differ in who decides the outcome, how long they take, and what happens if you stop.
| Dimension | How much you control | Realistic timescale | What it cannot do alone |
|---|---|---|---|
| Coverage of the subject | Almost entirely. It is a scoping and production decision. | A quarter or two for a genuinely bounded subject. | Distinguish you. Everybody can write the same set of pages, and increasingly does. |
| Connection between pages | Entirely, and it is the cheapest input on this list. | An afternoon once the pages exist, plus a habit afterwards. | Compensate for pages nobody needed. Linking filler to filler produces linked filler. |
| Demonstrable expertise | Partly. You control who writes and whether they are identifiable; you cannot fabricate knowledge. | Immediate if the expertise exists internally, otherwise not available at any price. | Substitute for coverage. One brilliant page about a subject is not a subject. |
| Being referenced by others | Barely. You control what you publish and whether it is worth referencing, not the decision to reference it. | Unpredictable, and rarely fast. Usually the slowest input by a wide margin. | Be bought safely. The market for buying it exists and carries a liability you do not control. |
Our position
How we treat this idea in a plan
What we do
- Describe topical authority as a model that predicts things, and separate what is documented from what is inference, in writing.
- Draw the subject boundary before any page count is discussed, and write it as a sentence that excludes things.
- Build the plan from questions people actually ask, and record what we removed and why.
- Design the internal linking as part of the plan rather than as a task after publication.
- Say plainly when a subject requires knowledge the business does not have, and recommend a different subject instead.
- Report progress as query breadth and position within the subject, both observable in your own Search Console.
What we will not do
- Present a keyword-generated map as a strategy, or bill for nodes nobody can justify.
- Claim that Google operates a topical authority score, or that a third-party authority metric measures one.
- Recommend publishing volume as a route to competing in a subject the business knows nothing about.
- Quote a number of pages required for a subject, because no honest number exists.
- Treat being referenced by others as something a content plan can deliver on a schedule.
Questions
Questions from people about to commit a year to this
Is topical authority a real ranking factor?
There is no documented system by that name, and anybody presenting it as a confirmed ranking factor is overstating what is known. It is a practitioner model that attempts to explain an observed pattern: sites that cover one subject thoroughly often compete better within it than sites that cover it once.
What is documented is adjacent and worth reading directly. Google’s guidance on helpful content asks whether a site has a primary purpose or focus, and whether content is produced by people with genuine knowledge of the subject. That is not the same claim, but it is the closest published statement to it.
How many pages does a subject need?
There is no number, and any figure quoted at you was invented. Coverage is finished when a reader with a reasonable question about the subject finds a page that answers it fully, not when a page count is reached.
The practical way to size it is to write out the questions first. List every question a buyer, a beginner and a sceptic would ask about the subject, remove the ones that belong on the same page, and what remains is your number. It is usually smaller than a keyword export suggests and larger than the three pages you were planning.
Should we buy a topical map?
Only if it comes with a stated reason for each node and a named question each page answers. Maps generated from keyword exports contain many nodes that exist because a phrase exists rather than because anybody wants that page.
A useful test before commissioning one: pick five nodes at random and ask who would search for that, what they want, and what they would do next. If the supplier cannot answer for all five, the map is a list with better graphics.
Does publishing more make a site more authoritative?
Not by itself, and volume is the most common substitute for depth. Forty unconnected pieces about forty subjects leave a site without a subject; eight connected pieces that exhaust one question can put it in contention for that question.
There is also a downside to volume that rarely gets mentioned. A library where a large share of pages have never been read is harder to maintain, harder to navigate and harder for anything to interpret than a smaller library that is complete and current.
Can we build authority in a subject we are not actually expert in?
You can build coverage. You cannot easily build the thing coverage is a proxy for, which is having something to say that a reader could not get from summarising other pages.
This is the honest limit of the model. Depth assembled from the first page of search results is a summary of what already ranks. The material that distinguishes a site comes from somewhere the competition cannot reach: your data, your operational reality, your customers, or people in your business who genuinely know the answer.
How would we know it is working?
Watch the queries rather than a score. The early signal is breadth: the number of distinct queries where you receive impressions inside the subject, and how many of them are questions you did not explicitly target.
After that, watch position on the harder terms in the same subject, and watch whether new pages on that subject start appearing faster than old ones did. Both are observable in Search Console and neither requires buying a metric from anyone.
Draw the boundary before you commission the pages
The decision that determines whether this work pays is the one made first: which subject, and what is excluded from it. That is a conversation, not a deliverable, and it is worth having before a map is bought from anyone.
Related
Where to go next
- if you already cannot reach the competitive termsThe diagnostic version: what to do when a competent site has hit a ceiling.
- finding the questions a subject contains
- consolidating what you have already published
- deciding what not to publish
- the part you do not control
- how this fits a wider search programme
- realistic timescales for any of it
Last updated