Does anchor text influence AI citation? A genuinely open empirical question

Does anchor text influence AI citation? A genuinely open empirical question

New terrain, and I want to be careful: most of what's said about anchors and AI Overviews / LLM citations is speculation dressed as insight.

The plausible mechanism: retrieval-augmented systems lean on the same web graph and may use anchor text as one relevance signal when selecting passages to cite, inheriting decades of anchor semantics from classic IR. If so, descriptive anchors pointing to a source could nudge its odds of being retrieved and cited.

The counter-mechanism: modern retrieval is increasingly embedding-based, matching query vectors to passage vectors, where the anchor string is a minor or absent feature compared to on-page content. Under this view, anchors barely touch citation selection.

On one hand, early observations that frequently-linked, descriptively-anchored pages appear often in AI answers hint at residual anchor influence. On the other, those same pages are simply authoritative and content-rich, which embeddings would favor regardless of anchors — the classic confound, now in a new venue.

Limitation: AI citation systems are opaque, change weekly, and we have no stable dataset to study, so any claim here is provisional by construction.

Practical read: don't restructure an anchor strategy for AI citation on current evidence; the signal, if real, is unquantified.

Open question: in retrieval-augmented generation, is anchor text a live relevance feature, or has dense retrieval rendered it nearly vestigial for citation selection?