<b>Anchor-ratio priors do not transfer across languages</b>

<b>Anchor-ratio priors do not transfer across languages</b>

Almost every published anchor benchmark is built on English-language profiles, and the resulting ratio priors get applied globally as if anchor behavior were language-invariant. It is not, and the reasons are structural rather than algorithmic.

Linking conventions are culturally and linguistically specific. Some languages and link cultures favor naked URLs; others favor descriptive phrases; in some markets the dominant CMS or forum software auto-generates a particular anchor form for every citation. The 'natural' baseline distribution is therefore a moving target across markets.

— An exact-match share that reads as aggressive in one language's organic baseline may be unremarkable in another's.
— A branded-anchor norm of 50 percent is an artifact of English-web citation habits, not a universal constant.

On one hand, the underlying engine is presumably one system applying consistent logic. On the other, that system is trained on observed natural distributions, which means its notion of 'natural' is itself language-conditioned — so the threshold that matters genuinely differs by corpus.

Limitation: I am aware of very few rigorous non-English anchor distribution studies, so the cross-language priors we'd need simply have not been published at comparable scale.

Open question: is there any defensible universal anchor benchmark, or must every market re-derive its own organic baseline from local profiles before any audit means anything?