<b>The "1-2% exact-match" rule is survivorship bias wearing a lab coat</b>

<b>The "1-2% exact-match" rule is survivorship bias wearing a lab coat</b>

You've seen the figure: keep exact-match anchors under ~2% of your profile and you're safe. It circulates as if measured in a controlled trial. It wasn't.

The number comes from observational studies — typically scraping the top 10 for a query set and averaging anchor distributions of ranking pages. A 2020 correlation study across several thousand SERPs reported median exact-match ratios in the low single digits for top results. Tempting to read as a target.

The flaw: these are <i>survivors</i>. You measure pages that already rank, then infer the threshold that let them rank. But you never see the pages with 2% exact-match that <i>didn't</i> rank for unrelated reasons, nor the high-ratio pages on domains strong enough to absorb it. The distribution of survivors is not the distribution of the safe.

On one hand, the central tendency is real — natural profiles genuinely skew toward branded and naked-URL anchors, so very low exact-match is a true property of organic links. On the other, treating a population median as a per-site ceiling ignores that tolerance scales with overall authority and link diversity.

Limitation: I've never seen this published with a control arm, because nobody can A/B-test a penalty ethically at scale.

Open question: is exact-match risk better modeled as an absolute ratio, or as a ratio conditioned on referring-domain count and trust?