Most clustering tools group the wrong thing

Give a typical clustering tool a keyword list and it groups by lexical similarity: terms sharing words go in a bucket. It is fast, it looks tidy, and it frequently produces groups that no search engine agrees with.

"cheap running shoes" and "running shoes sale" share two words and are usually the same intent. "seo audit" and "seo audit checklist" share two words and are usually different: one wants a tool, the other wants a document. Word overlap cannot see that difference. The results page can.

Cluster by SERP overlap instead

The reliable method is to ask which URLs actually rank for each query. If two queries return substantially the same results, Google considers them the same job and one page can serve both. If they return different results, they need different pages no matter how similar the words look.

This flips the workflow. Instead of grouping keywords and hoping, you group by observed behaviour and get a defensible answer to the only question that matters: one page or two?

The problem clustering is really for

Clustering exists to prevent two things.

Cannibalisation. When several of your pages target one query, Google picks one and discounts the rest. The picked page is often not your best. The symptom is a query where the ranking URL keeps changing week to week, or several near-identical pages splitting impressions.

Orphaned intent. A cluster with real search demand and no page assigned to it is a content gap, and it is far better evidence than any keyword difficulty score.

How UtilitySEO does it

Keyword clusters are built from your Search Console data by grouping queries according to which page Google actually routes each one to. That is the SERP-overlap method applied to your own site, using data Google gives you rather than scraped estimates.

Cannibalisation detection then surfaces the pages competing for the same query without any manual grouping, which is the step most teams never get around to doing by hand.

What to do with a cluster once you have it

Pick the strongest page as the target. Consolidate the near-duplicates into it, or redirect them if they add nothing. Point internal links at the winner. Then check the cluster again in a month: if the ranking URL has stabilised, the cannibalisation is resolved.