| Direction | Prompt Pattern | System Prompt Instruction | Core Measurement |
|---|---|---|---|
| Entity-to-Brand (E2B) | List ten brands that you associate with {entity}. |
Output ONLY brand names, one per line. No explanations. | Measures organic share of voice, top-of-mind brand recall, and competitive ranking for specific generic product or industry entities. |
| Brand-to-Entity (B2E) | List ten things that you associate with a brand called {brand}. |
Output ONLY items, one per line. No explanations. | Measures semantic depth, uncovering the core products, attributes, and topical associations the model directly links to an owned brand. |
| Query-to-Brand (Q2B) | List ten brands that you associate with the following search query: {query}. |
Output ONLY brand names, one per line. No explanations. | Measures commercial brand salience and retrieval positioning across granular, high-intent user search queries. |
These standardized probe templates strip conversational preamble to return clean, ordered lists that feed directly into analysis engines like the Treewalker framework for token-level logprob extraction or Matryoshka hierarchy attribution. Repeated across models and ungrounded execution rounds, they establish the exact baseline parametric footprint an LLM retains about an entity before any search tools or web context are introduced.
Referenced by
