The synonym-spin trap: diversity that isn't diverse

The synonym-spin trap: diversity that isn't diverse

A sophisticated failure mode. Aware that exact-match is risky, practitioners spin one keyword into 40 near-synonyms — "best running shoes," "top running shoes," "running shoes review," "buy running shoes" — and report high anchor diversity. But every variant is a transparent paraphrase of one commercial intent, and stemming/embedding-based analysis collapses them back to a single cluster.

On one hand, surface-form diversity is genuinely better than literal repetition. On the other, intent-level concentration is what distribution analysis can measure: semantically clustered anchors all map to the same target intent, so the "diversity" is cosmetic.

The deeper tell is unnaturalness. Real editorial links almost never produce 40 minor commercial variants; that pattern is itself a fingerprint of programmatic acquisition.

The fix:
— Cluster your anchors by embedding similarity, not exact string match, before measuring diversity
— Count topic-intent clusters, not unique strings; aim for genuine variety including off-keyword, descriptive, and partial anchors
— Accept naked-URL and "click here"-class anchors as legitimate diversity, since they occur naturally

Limitation: we don't know Google's exact clustering granularity, so embedding similarity is an approximation of an unknown internal model.

Open question: does commercial-intent classification operate at the string level or the semantic-cluster level when assessing anchor over-optimization?