Anchor risk and source quality interact — the same anchor is not the same signal

Anchor risk and source quality interact — the same anchor is not the same signal

We score anchors as if the text were the whole variable. But the identical anchor string carries different meaning depending on where it sits, and the interaction may dominate the main effect.

Consider "best forex broker" as anchor text. From a respected financial publication's editorial review, it reads as a legitimate descriptive citation. From a comment-spammed forum or a PBN footer, the same string reads as manipulation. The anchor didn't change; the conditional probability that a human chose it editorially did.

The hypothesis: Google's classifier effectively computes P(anchor | source), and only anchors that are improbable-given-a-quality-source raise risk. This explains why authority sites seemingly "get away" with exact-match anchors — on a trusted source, an exact-match anchor is plausibly editorial.

On one hand, this is consistent with the durable observation that link quality swamps anchor composition in practice. On the other, it makes anchor ratio almost meaningless in isolation, which contradicts how the entire tooling industry reports it.

Limitation: source quality is itself a Google-internal estimate we can only proxy crudely (DR, traffic, manual judgment).

Practical read: never evaluate an anchor without its source; a clean ratio from sketchy sources beats nothing.

Open question: is over-optimization fundamentally an anchor-source interaction effect, with the standalone anchor ratio a near-useless marginal?