seo-cluster

seo-cluster performs SERP-based semantic topic clustering to plan content architecture. Rather than grouping keywords by text similarity, it clusters them by how Google actually ranks them, measuring the overlap of shared URLs in the top-10 organic results for each keyword pair. The output is a hub-and-spoke content plan: a pillar page surrounded by spoke posts organized into subtopic clusters, complete with an internal link matrix and an interactive HTML cluster map.

The workflow runs in six steps. It begins by expanding a seed keyword or URL into 30-50 variants using WebSearch, mining related searches, People Also Ask questions, long-tail modifiers, and intent modifiers. It then runs SERP overlap analysis on candidate keyword pairs, applying thresholds: 7-10 shared results merge into a single post, 4-6 join the same cluster, 2-3 warrant cross-linking, and 0-1 stay separate. Keywords are classified by intent (informational, commercial, transactional, or navigational), with navigational terms excluded. The architecture step selects a pillar keyword, groups spokes into 2-5 clusters, assigns content templates by intent, and sets word-count targets of 2500-4000 words for pillars and 1200-1800 for spokes. A bidirectional internal link matrix is generated as a JSON adjacency list, and an interactive cluster-map.html visualization is produced from a template.

Use it when planning pillar pages, hub-and-spoke structures, or semantic keyword groupings for a content site. It can import keywords from an existing /seo plan output via --from strategy, and can optionally execute content creation through the claude-blog skill or output content briefs. If the DataForSEO MCP is available, it uses that for SERP data instead of WebSearch, with a cost-check step before each batch; otherwise it falls back to WebSearch. The methodology emphasizes avoiding keyword cannibalization by ensuring no two posts target the same primary keyword.