AIB · Artificial Intelligence Bureau
Methodology
AI answers are stochastic. One prompt × one run is not a measurement. Visibility is a distribution: repeats, paraphrases, language as authored, citations, position, competitors.
AI visibility methodology
What AIB measures
- Prompt families (not the raw prompt list — that is not the product)
- Modes: memory, grounded, retrieval, consensus
- Model identity: requested vs actual model, search engine, search_calls
- Two master numbers: AI Memory Score and AI Live Presence Score
- Confidence from observation budget, not from marketing adjectives
How the Bureau works
- A planner selects cells in a measurement cube. We never run the Cartesian product.
- Hebrew probes for Israel are authored in Hebrew and sent as-is.
- Scoring weights are profile-based. The 0–100 Presence Score is UX; the matrix is the asset.
- Historical comparison requires the same scoring version and comparable prompt families.
What this is not
- We do not publish proprietary prompt text.
- We do not equate API output with consumer UIs.
- We do not treat correlation as causation without a designed before/after.
Questions
What is Pulse confidence?
LOW. Pulse is 3 unique prompt types and about 15 model observations.
What are the default Presence weights?
Discovery 20%, recommendation 15%, share of voice 15%, citations 15%, accuracy 15%, consistency 10%, own-domain 5%, sentiment 5%. Profiles may override.
How often can information change?
Whenever a model, index, or page changes. Monitor plans rescan; a single Pulse is a snapshot.