Why a 30% exact-match ratio is meaningless on 12 links

Why a 30% exact-match ratio is meaningless on 12 links

The most common analytical error in anchor audits is computing percentages on samples too small to be stable. A profile with 12 backlinks showing "4 exact-match" reports 33% — but the confidence interval on that proportion (Wilson, 95%) spans roughly 14% to 61%. You are reading noise as signal.

On one hand, ratio thinking is correct: over-optimization is fundamentally about distribution, and Penguin-class scrutiny operates on proportions, not counts. On the other, a proportion is only informative when the denominator is large enough that resampling wouldn't flip your conclusion.

The mistake compounds when practitioners chase a target ratio ("keep exact-match under 5%") on thin profiles, then over-correct by buying branded links to dilute — manufacturing exactly the unnatural velocity pattern they feared.

The fix:
— Don't act on ratio targets below ~50 referring domains; report counts and a confidence interval instead
— Segment by acquisition cohort; a 5% lifetime ratio can hide a recent 40% spike that matters far more
— Treat the trailing-90-day anchor mix as the diagnostic, not the lifetime aggregate

Limitation: public link indices undercount, so even your denominator is an estimate with its own error.

Open question: does anchor-distribution scrutiny weight recent acquisition windows more heavily than lifetime mix, and if so, over what half-life?