GEO Citation Measurement¶
What It Is¶
The measurement loop for GEO. Because no Search Console exists for AI citation, the factory runs its own: a fixed prompt panel (from geo-prompt-research) executed across engines (ChatGPT, Claude, Perplexity, Google AI, plus any locally-dominant engine) on a schedule, recording for each prompt whether you're cited, where in the answer, and which competitors appear.
Metrics: AI Citation Frequency (AICF) / Share of AI Voice (your citations ÷ category citations), brand coverage rate (% of target prompts you appear in), and citation gap (prompts competitors own that you don't). (2026 tools — Otterly, Peec AI, Profound — are options; the metrics are the durable part.)
Why It's a Small-N Experiment¶
Citation is probabilistic: the same prompt fans out differently run-to-run and returns different sources, so a single check is noise — squarely minimum-detectable-effect territory. To measure validly:
- Sample each prompt multiple times to estimate a citation rate, not a one-off yes/no.
- Decide with the bayesian-decision-rule (posterior over citation rate), seeded by a marketing-benchmark-prior so thin samples don't whipsaw.
- Pre-register the lift that would justify a content change; don't over-react to one run.
How It Applies to Marketing Factory¶
This is the experiment-loop applied to GEO — it closes the factory loop by feeding measured citation gaps and wins back into geo-prompt-research prioritization, and it tells geo-content-pipeline which tactics actually moved citation rate. Running and parsing the prompt panel is fully agent-ownable; in a non-English market the panel must run in-language because citation behavior differs.
Related Concepts¶
- geo-prompt-research — supplies the panel and receives the feedback
- experiment-loop — citation measurement is its GEO instance
- bayesian-decision-rule — how to decide on noisy, probabilistic citation data
- llm-search — the fan-out mechanics that make citation probabilistic
Referenced from: geo-factory-operations