Two sites share the same niche. One has published two hundred posts in eighteen months, on a schedule, never missing a week. The other has published thirty. The thirty-page site outranks the two-hundred-page site for most of the terms that bring in customers, and the person running it is not writing more or building more links. She is writing in a shape the other site never planned.
That shape has a name in the trade: topical authority. In plain terms, topical authority is the pattern where a site earns visibility across a whole subject because it covers that subject thoroughly and connects the pieces, not because it pushed out a high volume of loosely related pages. You build it by picking a topic you can realistically own, writing one broad page that frames the subject, surrounding it with focused pages that each answer one narrower question, and linking them so the relationship is clear to a reader and to a crawler. It is closer to a planning decision than a writing decision, which is why volume alone rarely reproduces it.
What topical authority means, and what it does not #
Start with the honest part. There is no public documentation from Google describing a single ‘topical authority score’ that you can raise like a thermostat. The term is an industry label for something observed, not a confirmed dial inside the ranking system. What search systems do tend to reward, in line with Google’s own guidance on helpful, people-first content, is coverage that truly satisfies the range of things a person wants to know about a subject, produced by a source that demonstrably knows it. Topical authority is the name practitioners gave to the visible result of doing that well.
So the useful reading is behavioral, not mechanical. When your site answers the obvious question, the follow-up question, and the awkward edge-case question a subject raises, and does it in a way real people finish reading, you tend to accumulate rankings across the cluster rather than on one lucky page. When you publish one strong page in a field of otherwise thin ones, that page can rank, but the site around it gives search engines little reason to treat you as a reference on the topic. The difference is coverage plus coherence, not word count.
Two things it is not. It is not a reward for volume, since a hundred shallow pages on a subject can signal the opposite, that the site mass-produces. And it is not permanent, because coverage that made you the reference two years ago erodes as the subject moves and competitors go deeper.
Pillar pages and cluster pages play different roles #
A cluster has two kinds of page, and they are not interchangeable. The pillar is the wide page, the one that frames the whole subject, defines the shared vocabulary, and hands the reader off to deeper pages for the specifics. A cluster page (sometimes called a spoke) is narrow, taking one question the pillar only touched and answering it completely. Picture a page titled ’email deliverability’ that explains the landscape in one screen, then a separate page that does nothing but explain SPF, DKIM, and DMARC records in depth. The first is the map. The second is one street, walked end to end.
| Dimension | Pillar page | Cluster page |
|---|---|---|
| Scope | Broad, the whole subject | Narrow, one question |
| Search intent | General or early-stage | Specific, often long-tail |
| Job | Frame and route | Fully resolve one need |
| Links out | To every cluster page | Back to the pillar, and to close siblings |
| Length driver | Coverage of subtopics | Depth on a single subtopic |
The common mistake is writing every page at pillar width. When ten pages all try to be the broad overview, they compete with each other for the same query instead of covering ten different needs, and none of them wins. Width belongs to one page. Depth belongs to the rest.
Plan the cluster before you write a word #
Most weak clusters were never planned, they were assembled after the fact from posts that happened to be related. Planning first is what separates a cluster from a pile. A workable sequence looks like this.
- Choose a topic you can own. Not ‘marketing’, which is a library. Something bounded enough that a small team could plausibly become one of the better sources on it, like ‘cold email outreach’ rather than ‘sales’.
- Map the questions, not the keywords. Write down every real question a person moving through this subject asks, from first contact to edge case. Autocomplete suggestions, the ‘people also ask’ boxes, and your own support inbox tend to beat a keyword tool used alone.
- Group questions into pages. Each cluster page owns a group of closely related questions. If two groups answer the same underlying need, they are one page, not two. This is the step that prevents overlap later.
- Name the pillar last. The pillar is whatever broad frame ties the groups together. Naming it after you see the groups keeps it honest, so it frames what you actually cover rather than what you wished you covered.
Notice the friction this front-loads. It is slower than opening a blank document and writing whatever is due this week, and the payoff does not show up in month one. It shows up around the point where a competitor who published twice as fast starts to plateau and you do not, because their pages overlap and yours divide the subject cleanly.
How the links map inside a cluster #
Linking is where a cluster either becomes legible to a crawler or stays a heap of separate pages. The pattern is simple to state. Every cluster page links up to its pillar, the pillar links down to every cluster page, and closely related cluster pages link to each other where a reader would genuinely want the jump. Done consistently, this signals to a search engine that these pages belong together and that the pillar is the entry point.
The part people overlook is the sibling links, not the pillar links. A reader who finishes the DKIM page often wants the DMARC page next, and a direct link between them does more for both the reader and the crawl path than a third link back to the pillar. The test for any internal link is whether a person mid-task would truly want to click it. Links added only to hit a quota read as noise, and they dilute the ones that carry meaning.
This is architecture, not link craft. The finer points of how to phrase the link text, and how varied it should be, are their own subject and their own decisions. The cluster question is only which pages point to which.
Where clusters go wrong #
The model is not free of downsides, and three failure modes show up again and again.
Overlap that turns into competition. When two cluster pages drift toward the same question, they split the signals that would have made one page strong, and they can trade places in the rankings week to week. Diagnosing and untangling that overlap is a discipline of its own. For the purpose of building a cluster, the prevention is the grouping step above, keeping one clearly owned question per page.
Thin spokes. A cluster page that exists only to fill a slot in the diagram, with three shallow paragraphs and nothing a reader could not find faster elsewhere, weakens the cluster instead of adding to it. A missing page is better than a hollow one. If a subtopic does not have enough substance to justify a page, it is a section of another page rather than a page of its own.
An orphan pillar. Sometimes the pillar goes live and the cluster pages never get built, or they get built and never link back. A pillar with no supporting pages is only a long post, and it carries none of the coverage advantage the structure was supposed to create. The structure pays off when it is fully built and connected.
There is also a plainer cost worth naming. Clusters take longer to plan and are harder to reshuffle once live, so for a very small site chasing a handful of transactional pages, the overhead can outweigh the benefit. The model earns its keep on subjects broad enough to have real depth.
How to tell whether it is working #
Topical authority does not come with a dashboard number, so you read it indirectly. The signal to watch is not the ranking of one page but the spread. Over months, are more pages in the cluster picking up impressions and positions for their own queries, and is the pillar starting to rank for broader terms it could not touch when it stood alone? Growth that shows up across the set, rather than on a single page, is the pattern that matters.
Two honest cautions. This is slow, and it is hard to attribute cleanly, because a rankings rise during the same window could owe as much to a competitor slipping, a broad algorithm update, or seasonality as to your structure. Treat a single month as noise and a consistent multi-month direction as signal. And resist the pull to judge the cluster by its best page, because the whole point was the set, so the set is what you measure.
Frequently asked questions #
How many cluster pages does a pillar need? There is no fixed number, and a count set in advance usually produces filler. The right number is however many distinct, substantial questions the subject holds, which might be four or forty depending on the topic.
Should the pillar target a high-volume head term? Often it does, because the pillar covers the broad subject that the head term names, but forcing a pillar onto a term the page cannot fully satisfy tends to backfire. Match the page to the intent rather than to the search volume.
Can a page belong to two clusters? It can, when a subtopic sits squarely between two subjects, though it is worth checking that you are not describing two overlapping pages that should be merged. Shared membership is fine, duplicated coverage is not.
Is this the same as writing long content? No. Length is a byproduct of covering a subject properly, not the goal. A tight cluster of focused pages tends to serve a subject better than one enormous page trying to hold everything at once.
If you take one thing from this, make it the order of operations. The sites that build topical authority almost never do it by writing faster, they do it by deciding, before the writing starts, which subject they mean to own and how the pages will divide and connect it. The writing is the easy part. The shape is the work, and the shape is what the thirty-page site had that the two-hundred-page site did not.