TOPIC CLUSTERS
SERP-OVERLAP CLUSTERING

The skill starts from a seed keyword and expands it into 30 to 50 variants using related searches, People Also Ask, long-tail modifiers, and intent modifiers. Then it runs the part most other tools skip: for every candidate pair, it pulls the top 10 organic SERP results for both keywords and measures how many URLs overlap.

Claude SEO's default review bands treat 7 to 10 shared URLs as a same-page review, 4 to 6 as a shared-cluster review, 2 to 3 as weak adjacency, and 0 to 1 as likely separation. These are vendor heuristics, not Google thresholds. The reviewer should consider locale, device, date, query features, audience, and page purpose before approving the hub-and-spoke plan.

Primary references: Google's SEO Starter Guide explains topical site organization, and Google's sitelinks documentation covers logical site structure and relevant internal links.

Six Capabilities

The skill bundles six things most teams stitch together from separate tools. SERP-overlap scoring is the core; the other five make the output usable for content production.

Commands

How SERP-Overlap Scoring Works

For each candidate keyword pair, the skill fetches the top 10 organic results (ignoring ads, featured snippets, and PAA) for both queries and counts the shared URLs. Concretely: if keyword A has top-10 URLs {a1...a10} and keyword B has {b1...b10}, the score is |A ∩ B|. It is a raw intersection count, not a Jaccard score. The count maps to these review bands:

For 40 keywords, full pairwise analysis would create 780 keyword-pair comparisons. The skill reduces that work by pre-grouping with intent guesses (4 groups of about 10 means 180 within-group comparisons) and cross-checking selected boundaries. A comparison may require more than one search request, while assumed long-tail pairs may require none. Estimate the actual fetch batch and obtain approval before paid calls.

Output Files

For one page at a time, use the SEO content brief workflow. Cluster planning is the broader architecture step, while a content brief turns one approved target into a research-backed page plan.

All outputs land in the current working directory:

• cluster-plan.json: machine-readable cluster plan with keywords, clusters, links, and meta
• cluster-plan.md: human-readable cluster plan summary
• cluster-map.html: interactive SVG visualization, no build step required
• cluster-briefs/: per-post content briefs, generated when claude-blog is not installed
• cluster-scorecard.md: post-execution quality report

DataForSEO Integration

When the paid DataForSEO MCP server is installed and credentialed, the skill can use serp_organic_live_advanced instead of WebSearch. A cost check runs before every batch via claude-seo run dataforseo_costs.py check <endpoint> [--count N]. Calls above the approval threshold pause for confirmation; blocked budgets fall back to WebSearch. Search snapshots vary by date, locale, device, and provider, and neither route guarantees keyword volume. See /seo dataforseo for setup.