Metrics reference
Use consistent definitions when reading dashboards, exports, and reports.
Metric definitions are useful only when their population and denominator travel with the value. SEO scores depend on completed checks and coverage. GEO metrics depend on the selected answers, prompts, models, competitors, markets, tags, and time range. Search performance fields depend on the connected Search Console property.
Use this reference when building exports, reports, API consumers, or stakeholder dashboards. Preserve counts and filters beside rates and averages, and open the underlying evidence when a metric drives a decision. Similar labels from other products may use different formulas and should not be assumed equivalent.
When implementing a metric outside Sight, copy the definition, grain, window, denominator, exclusions, and rounding behavior together. A dashboard that reproduces only the label can create silent disagreements with the product. Validation should compare a small set of known projects and answer populations, including empty and partial cases, before the external calculation is used for recurring reporting.
| Metric | Definition | Important context |
|---|---|---|
| SEO overall score | Weighted summary of completed audit categories for a scan. | Coverage and tier affect which evidence ran |
| Visibility | Share of tracked AI answers that mention the selected brand. | Prompt and model population |
| Share of voice | Share of mentions among the tracked competitive set. | Competitor configuration |
| Sentiment | Measured tone of brand mentions. | Model output and evaluation method |
| Position | Average observed brand position in ordered recommendations. | Only meaningful where an order exists |
| Mentions | Count of brand references in the selected answer set. | Can exceed answer count |
| Retrieval rate | Share of eligible answers that retrieved a source. | Domain or URL grain |
| Citation rate | Share of eligible answers that cited a source. | Citation extraction quality |
| Gap score | Relative opportunity based on competitor source evidence and your absence. | Threshold and selected competitors |
| Clicks, impressions, CTR, average position | First-party search performance fields. | Available only through Search Console |
Keep denominator context
Every comparison should retain its answer count, prompt count, model set, date window, competitors, filters, and any missing or degraded probes.
Store counts and population fields with every exported rate or average. This lets reviewers distinguish a real movement from a change caused by missing probes, a smaller sample, or different filters.