You can’t improve a cluster you can’t see. A handful of measures tell you whether a group of articles works as a cluster or is just a list, and each one points to a specific fix.
Orphans
What it is: articles that no other article links to.
Healthy: none.
If not: link them first — it is the biggest gain for the least work. See Orphan pages.
Links between articles
What it is: how many of the possible links between the cluster’s articles actually exist. In SEO tools you may see it called density.
Healthy: articles link to the pillar and to their closest relatives. Small clusters can be densely linked; in large clusters a lower share is normal, as long as nothing is isolated.
If not: turn existing mentions into links, and link related articles to each other. See Internal linking.
How the links are spread
What it is: whether links are spread over the articles or concentrated on a few. Tools may call it evenness, entropy or show a Gini value.
Healthy: one clear main article with more links than the rest, and no article left out.
If not:
– Too even, no main article — choose a pillar and link the cluster to it. See Pillar pages.
– Too concentrated — one page gets everything, the rest little: add links between the supporting articles.
Article quality
What it is: whether the articles themselves are in good shape: enough depth, structure with subheadings, images where they help, a title and description.
Healthy: no thin or outdated articles dragging the cluster down.
If not: update or merge the weakest articles. See Meta titles and descriptions.
Putting it together
A single score is useful for comparing clusters and choosing where to start — but the fixes come from the individual measures. Work in this order:
- orphans,
- a clear pillar,
- links to the pillar,
- links between related articles,
- missing articles.
In ClusterIQ
ClusterIQ gives each cluster a health score from 0 to 100, made of five parts: article quality (30%), no orphans (25%), links between articles (20%), even spread (15%) and topic breadth (10%). It also names the cluster’s shape — Balanced Mesh, Authority Pyramid, Weak Structure and others, or Mixed when no single pattern stands out. Improve this cluster lists the fixes in the order above.
Don’t chase the score
A number is a way to find problems, not a goal. A link that doesn’t help a reader isn’t worth adding to raise a score — and a small, focused cluster can be excellent with numbers that look modest.
Back to the overview: Topic clusters.