Subjective Impression

Subjective impression is the Princeton GEO paper's human-evaluated score for how visible and prominent a source feels within an AI-generated answer, collected by having human raters read the answer and judge a source's presence directly, as a complement to the automated PAWC metric.

Why it matters

The paper reports subjective impression alongside PAWC specifically because a source can score well on a mechanical word-count metric while still not feeling prominent to a human reader, or vice versa, and the two metrics together give a fuller picture than either alone.

Related terms

• Position-Adjusted Word Count (PAWC)

  Position-Adjusted Word Count, or PAWC, is the metric the Princeton and IIT Delhi GEO paper (arXiv 2311.09735) uses to score a source's visibility within an AI-generated answer, weighting how many of that source's words appear in the response and where they fall, with earlier-appearing content weighted more heavily.

• GEO-bench

  GEO-bench is the benchmark the Princeton and IIT Delhi GEO paper built to test and compare optimization tactics, consisting of 10,000 queries spanning 8 different domains, used as the standardized dataset every tactic in the study, from statistics addition to keyword stuffing, was measured against.