Keyword clustering

A keyword clustering workflow:

1. Start with a research dump — keywords plus intent labels and volumes.

2. Group by intent first. All informational queries together, all comparison queries together, all transactional queries together.

3. Cluster within each intent. Use SERP overlap (do queries A and B share at least N of the top 10 results?), embedding similarity (cosine distance on query embeddings), or both. Tools like Ahrefs, Semrush, and Indexly automate clustering using these signals.

4. Map clusters to pages. Each cluster gets one pillar page targeting the head term plus body content covering the long-tail variants.

5. Add prompt clusters. Conversational AI prompts cluster differently from typed queries — cluster them separately, then link the resulting AI-prompt clusters to the SEO clusters they neighbor.

The output is a topical content map, not a keyword list — which is what both Google and AI engines reward.

Keyword research surfaces the universe of queries. Clustering decides how to ship coverage.

A keyword research dump with 5,000 queries can't be acted on directly — it implies thousands of pages, most of which would be thin and cannibalize each other. Clustering compresses that list into 20–50 topical clusters, each mappable to one substantial page.

In practice, the two are sequential: research surfaces the queries; clustering structures the response.

Three concrete payoffs:

1. Avoids cannibalization. Multiple thin pages targeting near-identical queries split signals and suppress all of them. Clustering forces consolidation into one strong page.

2. Builds topical authority. Google's ranking systems and AI retrieval both weight topical coverage. A cluster of 10 well-linked pages on a single subject signals authority that a single page on the same subject does not.

3. Lifts AI citation rate. AI engines (ChatGPT, Claude, Perplexity, Gemini) cite from sites that demonstrably cover a topic in depth. Cluster-driven content maps lift citation share more than the same number of disconnected pages would.

Five practices for production-grade clustering:

1. Cluster by intent first, then topic. A "best CRM" comparison and a "what is CRM" definition belong to different clusters even if they share keywords.

2. Use SERP overlap as the primary signal. Two queries that share 4+ of the top 10 results should almost always live in the same cluster.

3. Layer in embedding similarity. Cosine distance on query embeddings catches semantically related queries SERP overlap misses.

4. One pillar page per cluster. The pillar targets the head term; body sections cover the long-tail variants. Internal links from the pillar to deeper sub-topic pages reinforce the cluster.

5. Refresh clusters quarterly. As Google and AI assistants reshape SERPs, clusters shift. Pages that used to cluster together can split; separate clusters can merge.