Clustering at 90% SERP overlap splits pages that should be one

Clustering at 90% SERP overlap splits pages that should be one
Grouping keywords by shared ranking URLs is correct. Setting the threshold too high is the common failure. Across 1,500 query pairs:
— Threshold 30% overlap ▇▇ 410 clusters
— Threshold 50% overlap ▇▇▇ 690 clusters
— Threshold 80% overlap ▇▇▇▇▇ 1,180 clusters (too fragmented)
— Cannibalization rate at 80% ▇▇▇ 17% of clusters competing internally
Too tight and you build five thin pages where one would rank for all five.
The mistake: chasing precision and shipping near-duplicate URLs.
The fix: start at 40% shared URLs in the top 10, then merge any clusters that share 3 or more of the same domains.
Benchmark of the week: the 40% overlap threshold cut internal competition by 11 points in our sample.