Skip to main content

​
Direct answer

Use the evolution views in Cockpit > Overview with a consistent scope, then inspect the answers behind any change. Describe the metric as changing after an event, not because of it, unless you have a separate causal method.

​
Data required

  • Completed answers in the current and comparison windows
  • The same prompt cohort, providers, models, countries, languages, topics, and tags
  • Raw numerators and denominators for each metric
  • A log of prompt, monitoring, competitor, and site changes
  • Representative answers around the observed change

​
Workflow

  1. In Cockpit > Overview, select the current period and note the active filters.
  2. Compare with an equal prior window when the interface supports it.
  3. Record Mention Rate, Citation Rate, Share of Voice, Coverage, and their raw components separately.
  4. In Cockpit > GEO Matrix, locate the topics and providers contributing to the change.
  5. Read matching answers in Monitor > Response Analysis before interpreting an aggregate movement.
  6. Check your change log for prompt, model, schedule, country, competitor-classification, or site changes during the same interval.

​
Interpretation

Product Mention Rate and Citation Rate are response-level. Product Coverage is prompt-level. Product Share of Voice counts mention occurrences. A change in any one metric does not imply the others changed, and a percentage can move because its denominator changed. Prefer a fixed prompt cohort for trend analysis. If the cohort changed, show the old and new populations separately or label the discontinuity. Treat changes from small samples as provisional.

​
Possible next actions

  • Test whether an observed decline is concentrated in one provider, model, topic, or country.
  • Review newly absent prompts and the underlying answers before proposing content work.
  • Annotate a site release or campaign, then compare a stable cohort in later periods.
  • Restore a prior filter or prompt cohort when a configuration change explains the apparent trend.
Each action tests an explanation; it does not establish causality.

​
Limitations

  • Model updates, retrieval behavior, and answer variability can change output without a brand action.
  • Added or removed prompts can create a trend break.
  • Stored answers reflect configured monitoring, not all AI interactions.
  • Overlapping or unequal time windows can distort comparisons.