The distribution mistake: averaging anchors across the whole site

The distribution mistake: averaging anchors across the whole site

Practitioners frequently compute one anchor distribution for the entire domain and declare it healthy. But over-optimization is evaluated per target URL, not per domain. A site can show a beautifully natural 2% exact-match average while a single commercial landing page sits at 70% exact-match — and that page is the one at risk.

The failure is aggregation hiding concentration. Your homepage and blog naturally attract branded and URL anchors, which dilute the domain average and mask the money page's spike.

Per correlation studies of penalized profiles, the per-page anchor vector — not the site-wide one — is the relevant unit. Manual-action documentation has historically targeted specific URLs, consistent with page-level evaluation.

The fix:
— Build the anchor distribution per landing URL and rank pages by their individual exact-match concentration
— Flag any single page where one commercial phrase exceeds ~20% of its referring-domain anchors
— Watch for the inverse: a page with 100% branded anchors and zero descriptive anchors can underperform on relevance

Limitation: correlation across penalized sites doesn't prove the threshold is causal; thresholds are heuristics, not published constants.

Open question: is anchor-text risk assessed strictly at the canonical-URL level, or does it propagate across a cluster of internally linked pages sharing intent?