What it is, and where it came from
The method is simple: you write one comprehensive pillar page on a subject, ten to twenty cluster articles around it on individual sub-questions, and each of them links back to the pillar while the pillar links to each of them. The promise is that the search engine will treat your site as a "topic authority" as a result.
This is one of the most widespread pieces of content strategy advice today. It is worth knowing where it comes from: a software company's 2017 marketing material, which illustrated the claim with a chart. Underneath that chart there was no sample size, no control group and no significance test. The number that has been circulating ever since — "twenty to thirty posts in a cluster" — comes from there, as guidance.
The source that is not evidence
This does not mean the method is bad. It means we do not know what it is worth, and anyone claiming they do know is probably passing on the same material.
A few analyses have appeared since, trying to back it up. The most frequently cited one looked at a few hundred sites, with an observational method and no significance value — and, the part that is methodologically most delicate, it validated its own metric, the one it sells, with itself. That is circular. Not false, just not evidence.
What the search engine says about it
Representatives of the largest search engine have spoken on the question several times, and the position is consistent: they recognise no signal called "topic authority". There is no measure in their system for how many articles you have on a subject.
That is an important distinction. A cluster structure may still help indirectly — more relevant pages, more internal links, better navigation — but not through a secret score. And it changes what is worth doing: if there is no score, then the goal is not the piece count, it is that every article genuinely answers a real question.
And in AI answers?
The most honest answer here is that there is no data. We went through the research available on the subject, and there simply is no measurement examining whether topical depth or the internal linking web affects appearance in AI answers. The literature review summarising the field lists page-level structural factors neither as proven nor as disproven.
So anyone saying "the topic cluster is the new GEO" is claiming something there is no evidence for. What there is evidence for in AI citation is of a quite different kind — a concrete answer, a concrete number, genuine freshness — and the How to Get Cited by AI Answer Engines post covers it.
So why do it anyway
Because a cluster structure has four benefits that do not depend on the search engine.
- The reader really does work through it. If one article raises a question and the answer is sitting right next to it, the reader stays. That is measurable in your own analytics.
- It forces completeness. Write down how many sub-questions belong to a subject and you see immediately what is missing. That is the biggest benefit of the planning.
- It avoids self-cannibalisation. Without a cluster you write three similar articles on the same thing, and none of them turns out well.
- It is reusable. A finished cluster becomes the newsletter series, the social calendar and the customer service knowledge base too.
How to build one
1. ONE SENTENCE on the subject you want to be the expert in.
Not "marketing", but "visibility in AI search for a small business".
2. A QUESTION LIST, not a keyword list. 20-40 questions, as a buyer says them.
The structure of the cluster comes from here.
3. GROUP the questions into 4-6 subtopics.
If one question lands in a subtopic, it is not a subtopic: it is a paragraph.
4. THE PILLAR answers the "what is this and why" question, completely.
NOT a table of contents: an article worth reading on its own.
5. EVERY CLUSTER ARTICLE answers one question, and
links to at least ONE OTHER cluster article as well, not only to the pillar.
6. THE PILLAR is updated as the cluster grows. This is the step most often skipped.
How big it should be
The number going round — "twenty to thirty posts in a cluster" — has no measurement behind it: it comes from the 2017 report, as guidance, not from a study. What practice says: you need as many articles as there are real questions in the subject that deserve an answer of their own. If that is six, then six. If it is twenty-five, then twenty-five.
There is a usable rule of thumb from the other direction: if you cannot give a cluster article a title that is one concrete question, then that article is probably a section of another article. This test is more reliable than any piece count — and it protects you from the filling up of the cluster becoming an end in itself.
Common misconceptions
- "A proven method." Its source is a 2017 company report with no measurement apparatus. Usable, but not proven.
- "The search engine rewards topical depth." According to the search engine's representatives there is no such signal in their system.
- "It counts in AI answers too." There is zero data on that — neither for it nor against it.
- "You need twenty to thirty articles in a cluster." That number is guidance from the 2017 report, not the result of a measurement. You need as many as there are real questions.
What to take away
The topic cluster is a good planning tool and a weak SEO promise. Do it — because it forces completeness, protects you from repeating yourself, and really is better for the reader — but not because you expect a score from it. And for writing the articles the brief is the next step: the Content Briefs That Work post walks through it step by step.