Hierarchical clustering on cosine of bag-of-words: turns a raw KW list into a pillar → clusters → leaves cocoon plan. To structure a silo before writing.
A semantic cocoon (a.k.a. topic cluster) is a tree-shaped architecture where a pillar page (broad, generalist) links to several cluster pages (mid-tail) which link in turn to leaf pages (long-tail). Topical authority bubbles up from leaves to pillar, internal equity flows down from pillar to leaves. It's the opposite of a classic WordPress category that only links to its child posts without bidirectionality.
The tool tokenizes your KWs (lowercase, stop-words removed) then computes cosine similarity between each pair. Greedy agglomerative clustering merges the closest pairs until the target cluster count is reached. For each cluster, the proposed pillar is the shortest KW (most generic) or the most central one (max sum-of-similarities). It's a heuristic — sanity-check it before publishing.
Rule of thumb: 1 cluster per 8-15 KWs. A 3-KW cluster is too narrow (no credible pillar), a 30-KW cluster is too broad (the pillar becomes brand-only). For 60-80 KWs, aim for 6-8 clusters. For 150-200 KWs, aim for 10-15. If you have lots of long-tail KWs, raise the count — each leaf deserves its slot.
This tool produces the plan, not the content. For each retained KW you need a serious semantic brief (expected entities, Hn structure, NLP-relevant terms) then a human writer. That's exactly what we operate on the Nautilinks network: we take a brief, write the editorial piece, publish it on a network media with a dofollow backlink.
The other half is writing 60 coherent articles with serious semantic briefs and publishing them on authoritative media. The Nautilinks network handles that.