/topic-clusters — Topic cluster architecture

Seed topic: $ARGUMENTS

Step-by-step guide and worked example: ${CLAUDE_SKILL_DIR}/references/cluster-mapping-guide.md. Evidence for the depth rule (siteFocusScore in the 2024 leak; Google's Topic Authority system, announced May 2023 for news): ${CLAUDE_PLUGIN_ROOT}/skills/seo-fundamentals/references/patent-signals.md.

Strategic rule: depth over breadth

Per-domain topical focus is a stored attribute; per-topic authority is computed per source. A mid-authority domain with 200 articles in one cluster tends to beat a stronger domain with 50 articles spread over 10 clusters, for queries inside that cluster. Before opening a new cluster ask: is there an existing cluster to deepen instead? Open a new one only if it can reach 20+ interlinked pages within ~6 months. If the domain already covers 4+ unrelated clusters, say so and recommend consolidation.

Process

1. Validate the seed — broad enough for 8–15+ subtopics; aggregate demand; the business has (or can build) expertise; not every subtopic dominated by mega-sites; fits the domain's topical core.
2. Discover subtopics (15–30) — autocomplete, People Also Ask, related searches, competitor hubs, community questions (forums, support), question/comparison/best patterns.
3. Filter and prioritize with the 8-factor opportunity score (0–100): volume 20% · difficulty 15% · current position 15% · content gap 15% · business relevance 15% · conversion potential 10% · freshness need 5% · trend 5%. Label High / Medium / Low.
4. Entity-cluster mapping — each page gets one primary entity (singular, Knowledge Graph where possible) and 5–8 related entities; shared entities across clusters reveal bridge pages and inter-cluster links. Entity-anchored clusters survive query reformulation and AI Mode fan-out better than keyword-anchored ones.
5. Organize — pillar (3,000–5,000+ words, broad keyword, TOC, quarterly refresh) → cluster groups → cluster pages (1,500–2,500 words, long-tail) → supporting content (FAQ, glossary, templates/tools, case studies).
6. Linking map — pillar → every cluster page (descriptive anchors); every cluster page → pillar; 2–3 sibling links per page; cross-cluster via shared entities; no orphans.
7. Calendar — pillar first, then highest-priority cluster pages, then supporting content, then the rest; 2–4 pages/month per active cluster; complete a cluster in 2–3 months.

Also cover when relevant

• Quick wins — positions 11–20; high-impression low-CTR pages (→ /meta-tags); missing descriptions; thin pages already ranking; internal links from strong pages to underperformers.
• Declining content — 20%+ traffic drop MoM, 5+ position drops in 30 days, falling impressions over 90 days, 12+ months without update → refresh (data, sections, examples, meta, links, schema).
• Programmatic templates — location ("<service> in <city>"), comparison ("<A> vs <B>"), integration, use-case pages: ≥60% unique content per page, unique meta, localized data, unique visuals, links to pillar and siblings; never publish city-swap duplicates.

Output

1. Cluster overview — seed, pillar keyword and entity, page count, aggregate demand (with source or "est."), timeline.
2. Pillar spec — keyword, intent, length, sections, TOC, links.
3. Cluster page table — title · primary keyword · primary entity · intent · volume · difficulty · priority · group · status.
4. Internal linking matrix.
5. Content calendar — week · piece · type · keyword · priority · effort.
6. Schema per page type (Article + Organization/Person with sameAs + BreadcrumbList; consistent author across the cluster).
7. siteFocusScore note — how the cluster fits the domain's topical core.

Next: /content-brief for the pillar, /internal-links for the current site, keyword-mapper agent for cannibalization.