GEA measurement starts with a monitoring set: the prompts, written in buyer language, that represent each cluster. Run them on a fixed schedule against each engine you care about and record the outcome for every run. Without a stable monitoring set you have anecdotes, not measurement.
The core metric is citation share: the percentage of monitored prompts on which you are named, by engine and by cluster. Track it weekly. A single run is noisy, engines are non-deterministic, so read the trend across runs rather than reacting to one absence.
Three qualifiers make citation share useful. Position within the answer matters: first-named brands are chosen far more often than third-named ones. Framing matters: being named as the budget option is a different outcome from being named as the recommended option. And confidence language matters: hedged mentions convert poorly and usually signal an evidence gap.
Next comes the downstream layer. Answer-engine traffic arrives with few referrers and little query data, so instrument it directly: dedicated landing paths, on-site self-reported attribution ("how did you hear about us"), and a clean split between assisted and last-click pipeline. Most clients undercount GEA badly because the default analytics view hides it inside direct traffic.
Set expectations with volume honesty. GEA sends fewer sessions with materially higher intent. Reporting it on session volume makes a successful programme look like a failure; reporting it on qualified pipeline per session usually makes it the best line in the deck.
Finally, close the loop from measurement back into planning. Every weekly read should produce three lists: clusters where you gained, clusters where a competitor displaced you, and prompts where nobody is named well. The third list is your next quarter's roadmap.