PromptSpark.ai Docs
Sign in
Browse documentation
Docs Metric glossary

Metric glossary

Use one source of truth for the product's main evidence, status, source, and performance terms.

Brand performance

Term Definition
Visibility Percentage of completed responses where the brand was mentioned.
Mention Rate Answer-level brand mention coverage; currently the same coverage as Visibility.
Share of Voice Brand mention count divided by all brand mentions in scope.
Sentiment Normalized 0–100 score based on sentiment when the brand is mentioned.
Avg Position Average first-mention order; a lower number means an earlier appearance.
Discovered brand A brand found in answer text but not deliberately configured as the tracked brand.

Source evidence

Term Definition
Source A URL observed in provider search/retrieval or as an explicit citation.
Retrieval Source evidence used or returned while forming the answer.
Citation A source explicitly referenced in the final AI answer.
Citation Coverage Citation records divided by all observed source records.
Domain Coverage Distinct cited domains divided by distinct observed source domains.
Gap answer Competitor-visible answer with no owned-domain source evidence.

Runs and status

Term Definition
Monitoring run One cycle that asks each active prompt on its selected AI providers and analyzes the completed answers.
Completed All required answer collection and analysis work reached successful terminal state.
Partially failed The run has reportable completed evidence and one or more failed executions.
Failed The run could not produce the required reportable result.
Provider The AI platform that answers a tracked prompt.
Model The provider-specific model recorded for an execution and available in filters where supported.

Rules for reading metrics responsibly

Metrics are comparable only when their scope is comparable. Always carry the project, date range, providers, markets, prompt set, filters, and completion status alongside the value. A percentage without its response count can exaggerate the importance of a small sample.

  • Read a metric together with its numerator, denominator, or supporting count where the interface provides one.
  • Treat provider-level results as separate evidence before relying on the aggregate.
  • Use lower-is-better logic for Avg Position; most other performance scores improve upward.
  • Do not add percentages such as Visibility and Share of Voice; they describe different denominators.
  • Distinguish retrieval from citation when planning source work.
  • Treat partially failed runs as reduced coverage and inspect which providers or prompts are missing.
  • Use repeated answers and sources to support a conclusion, then preserve representative examples for explanation.