SERP-overlap clustering merges 31% of pairs that semantic clustering splits

SERP-overlap clustering merges 31% of pairs that semantic clustering splits
Two ways to group keywords: by word similarity (semantic) or by how many shared URLs rank for both (SERP overlap). We clustered 2,000 terms each way.
— Semantic only: 'cheap flights' and 'budget airfare' grouped, but Google ranks different pages: ▇▇▇ false merge
— SERP overlap (3+ shared top-10 URLs = same page): caught those splits, merged unobvious pairs like 'crm' + 'manage contacts': ▇▇▇▇▇
— Disagreement rate between methods: 31%
— SERP overlap costs API calls per keyword; semantic is near-free
So what: cluster semantically to draft fast, then validate borderline groups with SERP overlap. Google's grouping beats a thesaurus every time.
Benchmark of the week: keywords need only 3 of 10 shared ranking URLs to safely target one page; below that, split them.