SpyderBot Documentation
Data Sources and Data Classes
Before trusting a number, identify what was measured and where it came from.
AI-response data
Examples include prompts, raw responses, mentions, recommendations, citations, sentiment, entities, and response positions.
AI-response data can answer what selected AI systems returned under the declared conditions. It cannot directly measure website visits, orders, or conversions.
Machine Traffic data
Examples include bot or automated requests, requested URLs, timestamps, status outcomes, actor identity, and access-policy context captured by a qualifying source.
Machine Traffic can show observed request activity. It does not prove reading, understanding, indexing, citation, recommendation, or training use.
AI Referral Traffic data
Examples include human sessions attributed to an AI platform, landing pages, engagement, configured goals, orders, and revenue when the required source is connected.
AI Referral Traffic can measure attributed human journeys under the selected source and attribution rules. Ordinary referral data does not identify the exact prompt or response.
Audit data
Audit data describes the tested condition of a URL, page, entity, or configuration at a particular time. It is diagnostic evidence, not continuous traffic measurement.
Report data
A report is a saved representation of the available source-product result for a stated scope and time. A report may differ from a later live view because of data cutoff, corrections, source updates, filters, or method changes.
Estimated data
Estimated data is modeled, extrapolated, or supplied by an approved external estimation source rather than measured from connected first-party data.
Every estimate should display or link to:
- an Estimated label;
- provider or method class;
- period and freshness;
- relevant geography or population scope;
- whether the value is modeled, extrapolated, rounded, or range-based;
- a limitation explaining that it is not first-party measurement.
Estimated and observed values should remain visibly distinct and should not be silently combined. A high-level estimate can be useful for orientation, but connected first-party data should be used for operational attribution and outcomes when available.
Evidence labels
| Label | Meaning |
|---|---|
| Observed | Captured directly from an eligible source |
| Calculated | Deterministically derived from declared eligible data |
| Estimated | Modeled or extrapolated rather than directly measured from the relevant first-party source |
| Verified | Independently confirmed for the stated identity or condition |
| Classified | Assigned to a documented category |
| Inferred | A supported interpretation derived from available signals |
| Unavailable | Required data, access, source, or configuration is absent |
| Insufficient | Some data exists but cannot support the requested interpretation |
A result may use more than one label. For example, an observed request may have a verified actor identity and a classified likely purpose.