# Metrics reference

> Exact definitions of every number in your report — AI Presence Score, coverage, visibility rate, win rate, and sentiment — and how each is computed.

- Section: Reference
- Updated: 2026-07-11
- Canonical: https://knitknot.ai/docs/metrics-reference/
- Publisher: KnitKnot, the AI Presence Management platform (https://knitknot.ai)

---

This page is the precise definition of every metric in KnitKnot. Each is computed once at scoring time and read everywhere, so a headline number and the evaluations behind it never disagree. Throughout, an **evaluation** is one scored AI response — one prompt run against one engine.

## AI Presence Score

A **0–100 composite** of how favorably AI represents you across a benchmark, written at scoring time. It folds together how prominently you appear, whether you win comparisons, whether what AI says is accurate, and how favorably you're framed. It is a composite, not an average of the metrics below — those measure single dimensions; the score weighs all of them together.

Read the trend, not the absolute. See [The AI Presence Score](/docs/ai-presence-score/) for how the dimensions combine and why run-to-run variance is normal.

## Coverage

A **categorical label on each evaluation** describing how present you are in that one response. One of five values, most to least prominent:

- **Primary** — you're the main subject of the answer.
- **Substantial** — you're discussed at length, but not the sole focus.
- **Peripheral** — you're mentioned with some detail among others.
- **Incidental** — a passing mention.
- **Absent** — you don't appear at all.

Coverage is per-response and categorical. It's the raw material the visibility rate is built from — it is not itself a percentage.

## Visibility rate

The **share of scored evaluations where coverage is not `absent`** — i.e. how often you show up at all. A fraction from 0 to 1, shown as a percentage.

One critical detail: your report computes visibility **organic-only**. Head-to-head prompts that name your company directly (e.g. "Compare X vs Y") are excluded, because an AI mentioning you only because the question forced it to isn't real presence. What's left measures how often you surface when a buyer asks an open question about the category.

## Win rate (W-L-T)

For head-to-head comparisons, each evaluation resolves to one outcome: a **win**, a **loss**, a **tie**, or **not compared** (the prompt wasn't a comparison, or AI made no call). **Win rate = wins ÷ decided**, where *decided* = wins + losses + ties. **Ties count in the denominator** — a hedged answer is not a win.

Two rules shape this:

- Outcomes come from AI's actual pick in the response, extracted per feature and overall — not a holistic guess.
- A recommendation of one of your own products over another is a **brand win**, not a loss. Scoring understands your product family. See [Subjects](/docs/subjects/).

Win rate is also available per competitor and per feature, over the same underlying outcomes.

## Sentiment

How favorably you're characterized when you do appear, aggregated to a **0–100 scale** (higher is more favorable). Sentiment is measured per response and rolled up; it's independent of coverage — you can be highly present with lukewarm sentiment, or briefly mentioned in glowing terms.

## How they relate

- **Coverage** feeds **visibility rate** — coverage is the per-response label, visibility is the fraction of responses that aren't absent.
- **Visibility**, **win rate**, **accuracy**, and **sentiment** are the dimensions the **AI Presence Score** composites into one number.
- Every metric drills to its evidence: score → evaluations → response text → claim → cited source. See [Runs and scoring](/docs/runs-and-scoring/).