AI Visibility Metric Definitions
Overview
This glossary provides the stable terminology, denominator rules, and interpretation guidance used across CiteWorks Studio's AI Market Discovery reports.
Individual reports may include a short local explanation of the metrics most relevant to that category, but this page is the authoritative reference for how the terms are intended to be read across the research program.
For collection, qualification, and coding procedures, see the AI Market Discovery Methodology.
Core Research Units
Prompt-Surface Observation
One source query tested on one AI/search surface together with the resulting answer and associated collection metadata.
Example: One query tested in ChatGPT and the same query tested in Gemini are two separate prompt-surface observations.
Brand-Relevant Observation
A prompt-surface observation that satisfies the benchmark's tracked-brand relevance rule and can proceed to commercial buyer-intent classification.
A citation or incidental brand reference does not automatically qualify unless the answer satisfies the benchmark's relevance criteria.
Qualified Benchmark Observation
A brand-relevant observation that also fits one of the benchmark's defined commercial buyer-intent clusters and enters the applicable public analysis denominator.
The current buyer-intent framework includes:
- Brand Recommendation;
- Pricing & Value; and
- Multi-Brand Comparison.
Qualified Surface Breadth
The number of tested AI/search surfaces that contribute at least one qualified benchmark observation during a measurement period.
Important: Qualified surface breadth is not the same as the number of surfaces tested. It is a post-qualification result.
Recommendation Metrics
Valid Recommendation Coverage
The percentage of qualified benchmark observations in which a tracked brand appears inside a recommendation list and satisfies the benchmark's recommendation-validity rules.
Formula:
`Valid recommendations for the brand ÷ applicable qualified benchmark observations × 100`
Example:
41 valid recommendations out of 87 qualified observations = 47.1% valid recommendation coverage.
Interpretation:
This metric describes recommendation coverage within the qualified benchmark universe. It should not be interpreted as the percentage of all raw AI responses in which the brand is recommended.
Top-Three Recommendation Rate
The percentage of the applicable qualified benchmark denominator in which a tracked brand appears among the first three valid recommended options.
Formula:
`Qualified observations where brand ranks 1–3 ÷ applicable qualified benchmark observations × 100`
This metric is useful for distinguishing broad recommendation inclusion from lower-position appearances.
Rank-One Recommendation Rate
The percentage of the applicable qualified benchmark denominator in which a tracked brand is the first valid recommended option.
Formula:
`Qualified observations where brand ranks first ÷ applicable qualified benchmark observations × 100`
Rank-one rate is a stronger preference signal than simple recommendation presence because it measures how often the brand is presented as the first choice.
Valid Recommendation Shortlist Share
The share of qualified benchmark observations that contain a recommendation shortlist satisfying the benchmark's validity criteria.
This is a benchmark-level metric describing how frequently the qualified observation set contains a valid recommendation shortlist, rather than a brand-specific coverage metric.
Presence and Sentiment Metrics
Raw Mention Presence Rate
The percentage of the applicable observation set in which the tracked brand is explicitly present, regardless of whether the mention qualifies as a valid recommendation.
Presence and recommendation coverage should not be treated as interchangeable. A brand can be mentioned without being recommended.
Net Sentiment Score
A normalized measure of coded brand sentiment within the applicable benchmark observations.
The production benchmark should maintain one centrally defined scoring scale and aggregation rule. If the active rubric uses values such as negative, neutral, and positive, the public methodology should state the exact numeric mapping and calculation method.
Sentiment should be interpreted alongside presence and recommendation rank. A high sentiment score based on very few observations can be less informative than a slightly lower score supported by a much broader recommendation footprint.
Modeled Value Metric
Modeled AI Authority Value
A modeled comparative measure intended to represent relative AI discovery opportunity within the benchmark framework.
It is not:
- measured revenue;
- attributable sales;
- measured website traffic;
- advertising media value; or
- a guaranteed commercial outcome.
The metric should be labeled as modeled wherever it is shown. The public methodology should disclose the principal inputs and assumptions required for a reader to understand how the value is intended to be compared.
How Percentage Movement Is Displayed
When comparing two rates, the difference is measured in percentage points.
For example:
- July valid recommendation coverage: 27.3%
- August valid recommendation coverage: 44.8%
- Movement: 17.5 percentage points
In compact tables or charts, CiteWorks reports this as:
Up 17.5 points
Do not label a percentage-point movement as a simple percent change. A move from 27.3% to 44.8% is a 17.5-percentage-point increase and approximately a 64% relative increase; those are different calculations.
Publication convention
Use:
- Prose: “increased by 17.5 percentage points”
- Tables/charts: “Up 17.5 points”
- Declines: “Down 8.0 points”
Avoid abbreviated technical shorthand in public-facing reports when the full wording or the simpler word points is clearer.
Counts and Denominators
Pair Rates With Counts
Where practical, recommendation rates should be accompanied by their numerator and denominator.
For example:
47.1% — 41 of 87 qualified observations
This is particularly important for lower-visibility brands, where one or two observations can materially change a percentage.
Name the Denominator
A percentage should never imply a broader universe than the one used in the calculation.
If valid recommendation coverage is calculated using qualified observations, the report should not imply that the rate represents all raw prompt-surface observations collected during the research run.
Use Stored Counts
Production reports should calculate public rates from exact stored numerators and denominators. Rounded public percentages should not be used to reconstruct source counts when the underlying data is available.
Metric Interpretation Framework
A single metric rarely explains the full competitive position. Brand analysis is strongest when several signals are read together.
| Question | Useful metric(s) |
|---|---|
| Is the brand appearing at all? | Raw mention presence rate |
| Is the brand being recommended? | Valid recommendation coverage |
| Is it being recommended prominently? | Top-three recommendation rate |
| Is it the first choice? | Rank-one recommendation rate |
| Is the recommendation favorable? | Net sentiment score |
| How broad is the eligible AI/search environment? | Qualified surface breadth |
| Is the benchmark producing valid shortlists frequently? | Valid recommendation shortlist share |
| What is the modeled comparative opportunity? | Modeled AI Authority Value |
The strongest conclusions usually come from the relationship between these metrics, not from any one number in isolation.
Metric Governance
Metric names and formulas should remain stable across reports and measurement periods.
If a definition or formula changes:
- update the definition on this page;
- assign an internal methodology/version change;
- determine whether historical data can be recalculated consistently;
- annotate affected longitudinal reports when necessary; and
- avoid presenting pre-change and post-change values as directly comparable without disclosure.
For broader versioning and interpretation rules, see Research Standards & Interpretation.