Metrics
What every number in your report means and how to read it.
Counted or judged
- Counted in code
- Same answers in, same numbers out: visibility, mentions, position, buyer stage, citations and run deltas. No model decides them.
- A model's read
- Judgment, labelled as judgment: sentiment, topics, market position and the market leaderboard behind Brand Ranking and share of voice.
The Overall Score and Competition blend both.
Metrics at a glance
| Metric | Range | Type |
|---|---|---|
| Overall Score | 0 to 100 | Blend |
| Visibility | 0 to 100% | Counted |
| Mentions | A count | Counted |
| Position and mention rank | 1st, 2nd...; 0 to 100 | Counted |
| Competition | 0 to 100 | Blend |
| Market position | 0 to 100 | Model's read |
| Sentiment | 0 to 100, 50 neutral | Model's read |
| Topic sentiment | Up to 10 topics | Model's read |
| Brand Ranking | #1 to #10 | Model's read |
| Share of voice | 0 to 100% | Model's read |
| Visibility by buyer stage | 0 to 100% per stage | Counted |
| Citations | Counts and shares | Counted |
| Run deltas | Change vs last run | Counted |
Overall Score
One number for every engine and question in the run: OVERALL SCORE on Overview, AI VISIBILITY in Execution History and SCORE on the brand page. The tooltip reads "Composite of visibility, position, sentiment and coverage."
Step 1: score each engine
Four inputs, blended for each engine. Visibility counts most, because being named comes first.
Per-engine score inputs
- Visibility (counts most): share of questions that name you
- Market position: how often and how early, a model's read
- Sentiment: tone, a model's read
- Share of voice: your slice of the leaderboard
Step 2: combine the engines
The engine scores are then combined into one number, in a way that keeps one outlier engine from swinging it. The method is the same every run.
Example, four engines
- Two or three engines
- Same method, on the engines that ran. On Free, that's ChatGPT and Gemini.
- Watch for
- Free and Growth scores use different engines, so don't compare them directly. The same goes for a Partial run, which scores only the engines that answered.
Visibility
The share of questions where you're named at least once: the plainest answer to "does AI mention us?"
- How it's worked out
- Per engine: answers that name you ÷ all questions × 100. Then combined across engines.
- Example
- ChatGPT names you in 13 of 20 answers (65%), Gemini in 11 of 20 (55%). Visibility is 60%.
- In the app
- The VISIBILITY tile, the Visibility row in Score Anatomy and "You appear in" on Prompts.
Mentions
Every appearance of a brand's name in an answer, and the base of every counted metric.
- What counts
- Your name and variations, and each competitor's. Overlapping names aren't double counted, so "Tarnwell City" counts once, not also as "Tarnwell".
- Not counted
- Product names alone. Add them as a brand name variation.
- Mentions vs visibility
- Mention Trends and Total mentions count repeats within an answer. Visibility counts each answer once.
Position and mention rank
Named first beats named fifth.
- Position
- Where you're named in an answer, next to the tracked competitors. 1st means you're named before any of them.
- Mention rank
- How early you're named across the answers that name you, from 0 to 100. Named first every time scores 100. Shown as Mention rank in Score Anatomy and Ranking position per run.
- Watch for
- AVG POSITION in the Competitors tab's Brand Ranking table is different: it's perceived, from the market leaderboard, with a dash when an engine didn't rank that brand.
Competition
Visibility adjusted for tone.
- How it works
- Your visibility, nudged up when the engines are positive about you and down when they're negative. Capped at 100.
- Example
- Two brands with the same visibility: the one the engines describe positively scores higher.
- In the app
- Competition in Score Anatomy; Market Competition per run.
Market position
How strongly you show up when you do, per engine. One of the four Overall Score inputs.
- How it works
- A model's read of how often and how early you're named, per engine.
- In the app
- Market position score on each engine's card in the per-run Competition section.
Sentiment
How positively the engines frame you, separate from how often.
- How it works
- Per engine, a model labels you Positive, Neutral or Negative with a score from 0 to 100.
- Across engines
- One overall label and score, plus each engine's.
- Check it
- Sentiment Drivers on Sentiments lists the verbatim strengths and concerns. Read them before you quote the number.
- Why a model
- Tone needs judgment. A keyword count misreads "not the cheapest, but worth it", which is positive.
Topic sentiment
Themes the engines raise about you, like battery range or ride comfort. Each gets a label, the verbatim quotes and your share of voice on it. A negative topic becomes a Sentiment fix; a clear lead becomes a Strength.
Brand Ranking
Where you place when the engines are asked who leads your market: what they already believe.
- How it works
- Each engine is asked who leads your market for this persona and region, based on what it already knows. Their answers are combined into one ranking of up to ten brands.
- In the app
- The BRAND RANKING tile (#3 of 10) and the Brand Ranking tables, with your row marked YOU.
- Watch for
- It's perception, not a fact-check. You can win real answers and still rank low if an engine's built-in picture is out of date.
Share of voice
Your slice of attention next to the competitors the engines name.
- Where it comes from
- The same market read as Brand Ranking, combined across engines. The column sums to 100%.
- In the app
- SHARE in Brand Ranking, the Share of Voice vs Sentiment chart and Share of Voice Trends.
- Watch for
- A rising competitor can shrink your share even when nothing about you changed.
Visibility by buyer stage
Visibility split into Awareness, Consideration and Decision.
- How it's worked out
- Per stage: that stage's answers that name you ÷ that stage's questions. Each stands alone, so they don't sum to 100%.
- Example
- Awareness 47%, Consideration 57%, Decision 81%. Buyers who ask by name find you; category askers mostly don't.
- In the app
- Visibility by Funnel Stage on Overview, stage tiles on Prompts and Buyer Journey per run.
- Watch for
- Awareness is usually lowest, since those questions are about the category, not you. A gap that widens run over run is the warning.
Citations
The sources the engines point to: whose content shapes what buyers hear.
- Where they come from
- The sources each engine actually cited. Only answers that used live web results have them.
- Citation share by owner
- Each URL is your domain, competitors (from your list) or third-party. Subdomains roll up:
shop.tarnwell.examplecounts astarnwell.example. - Top domains
- The domains cited in the most answers, with their share.
- New and lost sources
- Domains cited this run but not last run, and the reverse.
Run deltas
What moved since the monitor's previous run, shown as chips like "+6 vs last run": score and visibility change in points, rank change (up means you climbed), per-engine change, competitors that climbed or slipped and new and lost sources.
A first run has nothing to compare, so no deltas. Deltas also flag which recommendations are new.
Coverage
Which personas and products the monitor's questions cover. An untested persona needs its own monitor or a place on this one.
Per-run report labels
Runs opened from Execution History show an At-a-Glance Report Summary, one column per engine (ChatGPT as OpenAI), with these names:
| Per-run report row | What it is |
|---|---|
| AI Visibility | How often you're named, and how early. |
| Market Competition | Competition: visibility adjusted by sentiment. |
| Ranking position | Mention rank. |
| Analyzed prompts score | Visibility for that engine. |
| Sentiment analysis | Sentiment for that engine. |
Score colors
Score Anatomy and per-run cells use four bands. Every cell also shows its number.
- 90 and above: dark green
- 70 to 89: light green
- 50 to 69: yellow
- Under 50: red