Topic clustering takes a flat list of keywords and groups them by semantic similarity. The output drives your site structure: one pillar page per cluster, supporting articles for the long tail.
How clustering works
Builder runs embeddings on every keyword in your set, then runs k-means with a silhouette check to pick the right number of clusters. You can override the count if you want fewer, larger groups (good for small sites) or more, tighter ones (good for big editorial teams).
Output
- A tree view: one cluster per parent, with its child keywords listed underneath.
- A suggested pillar title for each cluster.
- A "duplicate risk" warning when two clusters overlap.