Skip to main content
A first monitoring run is a snapshot of the selected prompts and models. Treat it as an initial observation, not as a trend.

Check the scope first

Before interpreting a percentage, confirm:
  • the expected prompts generated answers;
  • the intended models, countries, and language were selected;
  • topics, tags, and funnel stages are assigned correctly;
  • direct competitors are classified as Direct rather than Indirect or ignored;
  • the date range includes the run you want to review.

Read the Overview

Open Cockpit > Overview. The five cards summarize different parts of the selected dataset:
  • Mention Rate and Citation Rate use only answers with a SELF or DIRECT denominator;
  • Share of Voice counts mention occurrences, not distinct answers;
  • Avg. Sentiment summarizes scored mentions of your brand;
  • Coverage works at the distinct-prompt level.
See Core concepts before comparing these values with an export or API response.

Inspect the evidence

Open Monitor > Response Analysis to read individual answers. Use the page filters to narrow the table to the prompts, models, or answers you want to check. Then use:
  • Monitor > Competitor Mentions to verify brand relationships;
  • Monitor > Source Explorer to inspect cited domains and pages;
  • Cockpit > GEO Matrix to compare a metric across providers and prompt groupings.

When data is sparse

An empty chart can mean that no answer matches the current filters, that a model did not return data, or that the relevant signal was not detected. Widen the date range and clear filters before changing the prompt set. If answers exist but do not address the intended question, revise ambiguous prompts in Workspace Settings > Prompts. Keep the old and new wording separate if you need to preserve historical comparability.

Build a trend

Use repeated runs with a stable prompt set, model selection, countries, and personas to build a comparable time series. Document configuration changes when interpreting a break in the trend.