<b>Anchors as query reformulations: the under-modeled function</b>

<b>Anchors as query reformulations: the under-modeled function</b>

There is a function of anchor text rarely discussed in over-optimization debates: anchors may act as a corpus of human-written query reformulations, teaching the engine what phrasings people associate with a page. This reframes the value of diversity from 'safety' to 'coverage.'

The lineage is plausible. If a system learns associations between pages and the language used to reference them, then a profile rich in varied real anchors is effectively a labeled training set of how humans describe that resource — including synonyms and intents the page's own copy never used.

— Under this view, diversity isn't merely risk-reduction; it expands the semantic surface a page can match.
— A profile of fifty distinct organic anchors may surface a page for queries its on-page text doesn't even contain.

On one hand, this gives a positive, non-defensive reason to seek anchor variety — coverage, not just camouflage. On the other, it predicts that spun or synthetic variety adds little, because it doesn't reflect genuine human query language.

Limitation: this is a hypothesis about how association signals might be used, supported by the existence of related patents but not by any public confirmation that anchors are weighted as query reformulations.

Open question: if anchors function partly as crowd-sourced query labels, is the highest-value anchor not the one you'd choose, but the unexpected phrasing a real reader would?