DocsMetrics
Metrics overview
The six core metrics MentionFlow computes, the rules they all share, and where to read each one.
Every number in MentionFlow comes from the same small set of metric functions. So a value means the same thing wherever you see it: a scorecard, a table, a report, or the API.
The six core metrics
| Metric | Scale | Answers the question |
|---|---|---|
| Visibility | 0 to 100 | How prominently do engines mention you? |
| Share of Voice | 0 to 100% | Of the mentions across your tracked set, how many are yours? |
| Presence rate | 0 to 1 | In what share of answers do you appear at all? |
| Average position | rank (1 = best) | When you appear, how early are you mentioned? |
| Citation share | 0 to 1 | Of cited answers, how many cite your domain? |
| Sentiment | 0 to 100 | When you're mentioned, how positive is the tone? |
Five of them are the headline cards on your Overview: Visibility, Citation share, Share of Voice, Sentiment, and Estimated impressions. They sit as five equal cards, not a ranked bank.
Rules every metric shares
- The window. Metrics use a rolling window. It anchors at your newest collected day, not the wall-clock "today". The dashboard default is 7 days. You can also pick 14, 28, 90, or 180 days, or a custom range up to 200 days.
- The denominator is every run in the window. A run where you weren't mentioned counts as a zero for visibility and presence. Absence is data. Each metric's page gives its exact denominator.
- Blending. Headline metrics blend only the eight search-grounded engines. Share of Voice included. That way a knowledge-only engine's empty citations can't distort them. Per-engine views still show every engine collected, unblended. See Data honesty.
- Confidence. A metric computed from fewer than five runs carries a low-confidence flag.
- Null is not zero. Sometimes a metric has nothing to compute from: no citations yet, no scored mentions. Then it is null. It renders as an em-dash, never a made-up
0. See Data honesty. - A colour is a verdict, so it has to be earned. A metric runs red, then amber, then green only where we can publish a boundary and you can check it. Visibility is cut against its own formula. Citation share against a measured par. Share of Voice against an even split of your tracked set. Sentiment stays uncoloured on purpose: 50 is neutral by construction, which is a fact, not a caution. Estimated impressions is a modeled count with no denominator, so it gets no colour at all. Colour is never the only cue. The band is written out in words beside the value.
Versioning
These definitions are METRICS_VERSION = 1. If we ever change how a metric is computed, we bump that version and mark the change on charts. We never silently rewrite your history. The API serves the same metrics on the same scales.
Related
- The full derivation of engine sampling and demand tiers: Methodology.
- How these appear in the API: API conventions and the Resource reference.
- The daily series behind the charts, over the API:
GET /api/v1/metrics/history(also theget_metrics_historyMCP tool).