Signal Briefs

Google’s dedicated generative-AI Search Console reporting turns a previously inferred behavior into something publishers can observe longitudinally: when URLs from a site are surfaced inside supported generative Search experiences.

exmxc has frozen the first dataset in the AI Retrieval Observatory, covering May 30 through August 29, 2026 across exmxc.ai and TrailGenic.com.

The denominator matters

TrailGenic recorded 2,011 generative-AI impressions against 15,010 Web Search impressions, an AI Retrieval Share of 13.4%. exmxc recorded 241 against 7,085, or 3.4%. More important than the cross-site difference is what happens inside each property when AI appearances are normalized against ordinary Search exposure.

On TrailGenic, a Baldy winter field log reached a 62.2% AI Retrieval Share and a three-shoe field comparison reached 45.3%. On exmxc, the sPEG definition reached 23.5%, while a much more widely surfaced Alphabet signal brief reached only 1.3%.

Working model

The baseline does not establish causality. It does motivate a testable proposition: original evidence may create information gain, while knowledge architecture may make that evidence easier to identify, retrieve, and combine.

Four hypotheses are now preregistered before the next observation: first-hand evidence will outperform commodity summaries; decision-useful comparisons will outperform broad explainers; structured knowledge architecture will correlate with retrieval breadth; and consumer versus institutional knowledge will continue to show different device contexts.

Why this matters

AI retrieval can occur without observable traffic. That means traditional click analytics may understate a new form of influence: knowledge entering an AI-mediated answer even when the human never visits the source site.

The Observatory will re-measure monthly and publish longitudinal findings when the accumulated evidence warrants an update. Read Baseline #001 and the full methodology →