Visibility
Successful responses mentioning the brand divided by all successful responses in the selected scope.
Every score should lead back to evidence.
AI answers change. Wolf records the prompt, engine, target market, timestamp, completion status, raw response, detected brands, and cited URLs for every check.
Integrity commitments
- Raw-answer traceability
- Stable prompt cohorts
- Source/citation separation
- Transparent formulas
- Failure-aware reporting
- No charge for failed checks
- Run metadata
- Exportable evidence
Wolf measures the responses collected for your defined prompt set. It does not claim to represent every possible answer an AI engine may show every user.
Collection method
What one observation contains
One Wolf check is one prompt executed against one selected AI engine for one supported regional or language target. A completed observation can include the prompt, engine, target, execution timestamp, completion status, raw response, detected brands, relative mention order, citations, and exact source URLs made available by the engine.
Brand detection turns unstructured answer text into comparable observations. Wolf distinguishes a brand mention from a citation: an engine may name a company without linking to its website, or cite a third-party source that discusses several brands. Keeping those concepts separate makes the resulting analysis easier to audit.
Checks that cannot be completed are recorded as failures and excluded from successful-response metrics. They are not charged. This prevents an unavailable engine response from being silently counted as a brand omission.
Share of voice
The tracked brand's mentions divided by all tracked-brand mentions in the same response cohort.
Relative position
The order in which a brand appears among detected brands when an answer contains an ordered recommendation.
Citation presence
Whether an available cited URL or domain belongs to the tracked brand or another relevant source.
Interpretation
How to avoid false confidence
Generative answers are probabilistic observations, not permanent search positions. They may change with model updates, time, location, language, personalization, retrieval behavior, and source freshness. A one-off response is useful evidence of what occurred, but it is not a census of every answer shown to every user.
Meaningful trend analysis uses a clearly defined prompt cohort, comparable engines and targets, and enough successful observations to understand variation. Teams should review the underlying answers when a metric changes and avoid attributing causation to content or PR work without supporting evidence.
If you are starting with one assistant, the ChatGPT rank tracker guide turns these principles into a 30-observation manual baseline with prompt patterns and explicit formulas.