Optimizing citations before retrieval is fixing the wrong problem.

A GEO plan can start with passage structure, then treat weak citation share as a selection problem. The causal mistake comes earlier: when an answer comes from trained memory, no citation slot opens. Passage rewrites cannot improve selection in a contest that never began.

Research maps retrieval to freshness, popularity, uncertainty, complexity and platform policy threshold. My read is that retrieval eligibility is AI visibility's missing denominator. Niche and long-tail facts may create citation opportunity because they are harder to answer from memory.

For example, imagine a niche supplier publishes a dated note on a revised standard while a category leader keeps evergreen copy. A current, long-tail query may require retrieval, giving the niche page an opportunity. Calling both selection failures sends effort to the wrong fix.

The research supports a clear operating rule: check whether target queries retrieve before changing passage structure or measuring citation share. If they do not, prioritise fresh, dated, specific, long-tail and corroborated material; if they do, compare selection quality within the retrieval-eligible set.

#AISearch

#GEO

How does your AI-search reporting separate retrieval eligibility from citation selection?