Access and context
Defined fields, fixed assessment rules and traceable sources.
CITARA INSTITUTE · REPORT
AI Recommendation Share-of-Voice Benchmark Method before ranking.
The AI Recommendation Share-of-Voice Benchmark is a planned Citara measurement system with a defined sample, fixed prompt or assessment framework, repeatable rules and visible limitations. Results are published only when the method and aggregation support responsible interpretation.
Direct answer
The AI Recommendation Share-of-Voice Benchmark is a planned Citara measurement system with a defined sample, fixed prompt or assessment framework, repeatable rules and visible limitations. Results are published only when the method and aggregation support responsible interpretation.
Author: Thorsten KellerUpdated: 2026-07-30
Measurement dimensions
What the report actually observes.
Recommendation visibility against a fixed prompt universe and clearly defined competitive set.
Recommendation and competition
Defined fields, fixed assessment rules and traceable sources.
Evidence and change
Defined fields, fixed assessment rules and traceable sources.
Methodology
Every release must explain how it can be repeated.
Date, models and versions, country, language, prompt set, sample, scoring, repeatability rules, limitations and data type are documented visibly.
| Field | Requirement |
|---|---|
| Research frame | Date, models, country and language |
| Sample | Definition, size and exclusion rules |
| Measurement | Prompt set or assessment, scoring and repetition |
| Transparency | Observed, inferred or customer-provided |
| Limitations | Uncertainty, coverage and aggregation threshold |
Privacy
No identifiable confidential customer cases in public benchmarks.
Sectors and groups below the minimum aggregation threshold are suppressed. Private customer documents, quotations and precise performance figures are not used publicly.
Frequently asked questions
Direct answers before the next step.
Is the report already published?+
The methodology page is live; results follow only with a stable method and sample.
Which models are tested?+
Each release lists models and versions where available.
How are prompts selected?+
Through a documented buyer, category or readiness logic.
How are small samples handled?+
Results below the aggregation threshold are not published.
Can the measurement be repeated?+
The prompt set, rules and time frame are documented for repetition.