Metrics reference
Understand EqualAI's actual denominators, rank-weighted visibility, and differences between reports.
View as MarkdownAlways identify the screen, project, date range, and active filters before comparing a metric. A familiar label can describe a different calculation on another screen.
Mentions and visibility
Brand mentions counts responses that name a brand. Repeating the name several times in one answer does not create several mentioning responses.
Mention rate is 100 × mentioning responses ÷ responses in scope.
AI visibility in Overview and AI Model Insights is an average rank-weighted score. For a response naming the brand at a valid rank, its contribution is round(100 ÷ rank). Rank 1 contributes 100, rank 2 contributes 50, and rank 3 contributes 33. An absent brand or missing/invalid rank contributes zero. Those zeros remain in the response denominator.
For example, three answers contributing 100, 50, and 0 have an average visibility of 50. This is an illustrative example. It is not a search-engine ranking or a percentage of all AI users who saw the brand.
AI Visibility Reports uses a different calculation: its visibility percentage is the rounded percentage of evaluated responses mentioning the brand. Do not equate this with the rank-weighted Overview score. See AI Visibility Reports.
Share of voice
Overview share of voice is 100 × a brand's mentioning-response count ÷ the sum of those counts for the project and tracked competitors.
One response can name several tracked brands and contribute to several brand counts. This is a comparison within the tracked set, not total industry market share. Changing the competitor set can change the denominator.
A change from 20% to 25% is +5 percentage points, not +5% relative growth. Share-of-voice movement uses percentage points.
Citations: check the screen
| Screen or metric | Numerator | Denominator |
|---|---|---|
| Overview citation rate | Responses citing the project domain | Responses in scope |
| AI Model Insights citation-rate matrix | Responses citing the row brand's domain | Responses for that model in scope |
| Responses citation rate | Responses containing any source citation | Responses in scope |
| Citations: Citation share | Citation occurrences for a source group | All citation occurrences in the filtered source set |
| URL Report citation rate | Loaded project responses matching the report target | All loaded project responses |
Citations can count multiple occurrences in one response. A citation rate counts a qualifying response once even when it has several qualifying citations.
Overview's project-domain matching normalizes the domain and includes matching subdomains. Do not assume every report applies domain settings identically; investigate the actual source URLs when exact subdomain isolation matters.
Sentiment
The Sentiment report assigns positive and very positive responses +1, negative and very negative responses −1, and neutral/unknown responses 0. Its net score is the rounded sum divided by all responses in scope, multiplied by 100. Unknown responses remain in that denominator; zero is not proof that every answer was classified neutral.
In AI Model Insights, the brand comparison sentiment matrix averages response sentiment among answers mentioning each row's brand. Competitor rows reuse the captured response sentiment; they are not independent competitor-specific tone evaluations. This differs from the own-brand model summary's all-response denominator.
Aggregation and missing evidence
- AI Model Insights matrix averages weight model columns equally; they are not necessarily pooled response averages.
- Some Overview brand charts carry the last measured value into empty time buckets. This does not create new observations.
- A model with little data can change sharply after a single answer.
- An unavailable rank, no captured citation evidence, and a verified absence are distinct underlying conditions, even when a summary renders zero.
- Ads, reputation dimensions, weekly summaries, and technical checks have their own scope. Follow their guides rather than substituting these formulas.