Semantic Clustering
Method
- Take seed keyword (e.g. "claude code")
- Fetch SERP for seed via WebFetch/Playwright (top 10 results)
- For each related keyword (autocomplete + "People Also Ask"):
- Fetch its SERP
- Compute overlap with seed's SERP (Jaccard index)
- Group keywords where SERP overlap ≥ 30% → same cluster
- Cluster center = highest-volume keyword
Output
# Cluster: "claude code"
## Pillar: claude code (vol: 12K, KD: 45)
- Intent: informational
- Featured: AI Overview, video
## Cluster pages
1. claude code installation (vol: 2.4K)
2. claude code vs cursor (vol: 1.8K)
3. claude code mcp servers (vol: 900)
4. claude code hooks (vol: 720)
Anti-Cannibalization Check
Before creating cluster pages, verify no existing page targets same intent. Use seo-content skill.