A topical cluster is a set of search queries that mean roughly the same thing to a search engine — different phrasings of a single underlying intent.
For example, all of these queries belong to one cluster:
- what is a data glossary
- data glossary definition
- how do I create a data glossary
- data glossary tool
A traditional SEO will call these "long-tail variants of a head term." BCited calls them members of a topical cluster.
How BCited discovers clusters
- Pull 90 days of ranking queries from Google Search Console
- Embed each query as a 1536-dimensional vector via OpenAI's
text-embedding-3-small - Run a centroid-greedy clustering pass over Cloudflare Vectorize: each unclustered query gets compared against existing centroids; close matches join the cluster, distant ones start new ones
- The LLM generates a human-readable name for each cluster from its top member queries
The result is typically 40–80 clusters per site — a manageable strategic surface instead of thousands of raw queries.