Topical clusters: grouping pages so a subject is visible
A cluster is a pillar page and the pages around it, linked so the group reads as one subject. The idea is sound and the diagram is drawn far more often than it is earned, so this guide is mostly about when not to restructure anything.
The short answer
A topic cluster is a hub page covering a subject broadly, a set of pages covering its parts in depth, and links running both ways between them. It is worth building when you have genuine depth on a subject and the pages already exist in some form. Building the structure first, and the content to fill it later, is the cargo cult version.
What a cluster is, stripped of the diagram
Three claims, and only the third is doing any work. First, that one page should cover a subject broadly. Second, that other pages should cover its parts in depth. Third, that the links between them should be dense and reciprocal, so that arriving at any page in the group gets you to the rest of it.
The mechanism underneath is not mystical. Internal links are how pages get discovered and how importance is distributed inside a site, and a dense group of links between pages about one subject means every page in it is reachable, contextualised by the others, and described by the anchor text of its siblings. That is all. There is no separate ranking system that recognises clusters and rewards them.
Which is why the useful form of this check is not "do we have clusters" but "are the pages about one subject connected to each other, or does each one sit in its own hole". Sites fail this by publishing forty articles over three years, each linked only from the blog index, none of them aware the others exist.
The two pages that test whether you have a cluster
The tidy version of this diagram is never the problem. The problem is always the page two groups both want, and the page no group claims.
For the page two pillars want, pick one and link from the other. A page with two parents is a page whose subject you have not decided, and it is the state most likely to end up competing with a sibling for the same query. Choose the pillar whose readers would search for it, put it there, and let the second pillar link to it as a related page rather than as a child.
For the page nothing claims, the honest options are to leave it outside the structure or to remove it. A seasonal gift guide is not part of a subject and does not need a parent. What it does need is a link from somewhere, because a page reachable from nothing is an orphan regardless of how the rest of the site is organised, and that is a real defect where cluster membership is a preference.
When the model is cargo cult
The pattern gets applied to sites that do not have the material for it, and the result is worse than no structure at all. Three signs you are in that situation:
- The pillar page is a table of contents. A page whose whole content is links to its children is not a pillar, it is a category index with aspirations. A pillar earns its place by being the page somebody would read if they only read one, which means it has to say something on its own.
- The children were commissioned to fill the diagram. Deciding a subject needs twelve supporting pages and then finding twelve titles is how a cluster of thin pages gets built. The pages should exist because somebody asks those questions, and a group of thin pages is still thin however neatly it is linked.
- The site has forty pages. Below a certain size the structure is the navigation. A small site does not need a cluster architecture, it needs its pages to link to each other where a reader would want that, which takes an afternoon and no diagram.
The corollary is that the reorganisation this check appears to recommend is rarely the right first move. Adding the missing links between pages you already have is cheap, reversible and captures most of the benefit. Rebuilding a URL structure into pillars and spokes is a migration, with redirects, and it should be justified by something more than a shape.
Clusters are a way of grouping pages, not a way of grouping URLs.
You do not need /sharpening/grits for a page to belong to the sharpening group. The grouping is made of links, and a flat URL structure with dense internal linking is a real cluster. Sites that believe otherwise end up restructuring their addresses, which costs redirects and gains nothing that a link would not have given them.
Where the URL path does help is human legibility, in a breadcrumb and in a shared link. That is a good reason to nest paths and a poor reason to migrate an existing site.
A grouping is a shape in the link graph
Clusters are not a property of any page. They are what the internal links look like once you have collected all of them, which is why this is the last check in this pillar that a single fetch could ever answer.
This one needs the full crawl
A cluster is a shape in the link graph, so it comes into existence only once every internal link on the site has been collected and the groupings can be seen.
The instant search on this site audits a single page, so rather than show you a verdict it cannot support, this guide sends you to the place the check actually runs.
Does this affect AI search?
Indirectly, and the honest version of the mechanism is less exciting than the usual claim. There is no system reading your site structure and concluding that you are an authority on a subject. Retrieval happens at the passage level, and a passage does not know which cluster its page was in.
What the grouping does affect is coverage. A subject treated across a pillar and eight linked pages produces many passages that answer many adjacent questions, all reachable, all discovered. The same eight pages published without links between them produce the same passages and a discovery problem, and a page nothing links to is a page that may not be in the corpus at all.
There is one more effect worth naming, and it is about internal consistency. When a group of your pages disagrees with itself, an old post saying one thing and a newer one saying another, a system summarising your site has to pick. Clustering does not fix contradictions, but it makes them visible to you, because reading a pillar and its children in sequence is how you notice that three of them give different answers.
Why restructuring a site is rated a Refinement
Because there is no failure state. A site with no clusters at all is crawled, indexed and ranked on the merits of its pages, and plenty of well-organised subjects are covered by sites whose structure is a flat blog. Nothing is excluded, no format is lost, and no threshold is missed, which puts it squarely at the bottom of the scale.
The rating is also a deliberate brake. This is the finding most likely to be turned into a quarter of work, and the work most likely to be a migration with redirects, changed URLs and a period of instability. Rated any higher, it would authorise a project that the evidence does not support.
The part of this check that genuinely matters is already rated higher elsewhere. If your pages are not reachable, that is an orphan finding. If they are thin, that is thinness. If they compete, that is duplication. Clustering is the tidy explanation of why those three keep happening together, and it is worth understanding for that reason rather than for its own.
Finding which of your pages are connected to nothing
You do not need a graph tool to get the useful answer. Count how many internal links each page receives from within its own section, which is the number that tells you whether a group exists.
SECTION=/guides/
# every page in the section, from the sitemap
curl -s https://example.com/sitemap.xml \
| grep -o '<loc>[^<]*</loc>' | sed 's/<[^>]*>//g' \
| grep -F "$SECTION" | sort -u > /tmp/pages.txt
# how many other pages in the section link to each one
while read -r page; do
path=$(printf '%s' "$page" | sed 's|https://example.com||')
hits=0
while read -r other; do
[ "$other" = "$page" ] && continue
curl -s "$other" | grep -q "href=\"$path\"" && hits=$((hits + 1))
done < /tmp/pages.txt
printf '%s\t%s\n' "$hits" "$path"
done < /tmp/pages.txt | sort -nIt is slow and it answers the question. Anything with a zero is connected to its own subject by nothing, which is the finding worth acting on. Anything with one is linked only from the pillar, which is a spoke and not a cluster.
Then read the top of the list. The pages your own section links to most are what your site actually treats as its pillars, and they are frequently not the pages anyone nominated.
Questions this check raises
- Do I need a pillar page and cluster content?
- Only if you have enough real material to fill both. The structure works because a hub concentrates authority and distributes it to pages that deserve it, which requires the pages to deserve it. Publishing a hub and twelve thin sub-pages to satisfy the diagram produces thirteen weak pages instead of one adequate one.
- How do I know if I already have a cluster?
- Look for the two links that define it: does the hub link to every sub-page, and does every sub-page link back to the hub? A group of related pages that do not reference each other is a category, not a cluster, and it gets none of the benefit.
- Is a topic cluster the same as a category?
- No. A category is a taxonomy label applied to pages; a cluster is a link structure with a page at its centre that is worth reading on its own. Most category archives are lists with no content of their own, which means they concentrate nothing and pass on nothing.