What Topic Cluster means
The pillar covers the topic comprehensively at a high level. Each spoke goes deep on one sub-topic and links back to the pillar, which links out to all of them.
The interlinking is the point. It declares which page is definitive for the broad topic and how the sub-topics relate to it.
Why it matters for AI search
Clusters help AI systems resolve which page owns which concept — the ambiguity that hurts most when several pages could plausibly answer.
They also map naturally onto query fan-out: the pillar answers the head question, spokes answer the sub-questions a model decomposes into.
See it in action
A pillar and its spokes, connected.
AI Visibility (pillar)
Crawler access
How AI crawlers reach your site
Entity resolution
Being a known thing, not a string
Answer structure
Writing extractable passages
Citation building
Earning third-party corroboration
Measurement
Benchmarking citation share
Platform differences
What varies across ChatGPT, Perplexity, Gemini
One pillar, six spokes, bidirectional links.
The structure tells a model that the pillar owns the broad concept and each spoke owns a sub-concept. Without the links it's just six unrelated pages.
How to get it right
Building topic clusters
- Define the pillar topic narrowly enough that you can own it
- Give each spoke one clear sub-topic with no overlap between them
- Link every spoke to the pillar and the pillar to every spoke
- Cross-link spokes where the relationship is genuine, not mechanically
- Resolve any existing pages competing with a spoke before publishing it
Common questions
How many spokes should a cluster have?
However many sub-topics genuinely exist. Six to twelve is common. Padding to hit a number produces thin spokes that weaken the cluster.
Is a topic cluster the same as a topical map?
A map defines the whole domain; a cluster is one pillar and its spokes within that domain. A map usually contains several clusters.
Do clusters still work?
The underlying principle does — comprehensive coverage with explicit relationships. The mechanical version, where the structure exists but the content is thin, never worked well.
Related concepts
These come up alongside Topic Cluster constantly.