Framework • Track 04
The Entity Measurement Framework
Rank tracking in LLMs is dead. An AI response is a sample drawn from a probability distribution. This framework operationalizes empirical prompt sampling: replacing arbitrary formulas with portfolio breadth, dual-portfolio tracking, and verified visibility vectors.
The Three Distinct Visibility Vectors
Stop collapsing visibility into a single vanity metric. Model interaction happens across three distinct behaviors:
Presence (Mention Rate)
Responses Mentioning Brand ÷ Total Sampled Runs
Did the model surface your brand at all? Mention Rate is our fundamental presence metric across sampled prompts. It proves existence, not endorsement.
Recommendation Rate
Recommended Instances ÷ Total Evaluated Runs
Did the model actively recommend your entity as a top viable solution, or was it merely referenced as an alternative or historical footnote?
Citation Share
Domain Citations ÷ Total Retrieved URLs
Did the model pull your specific domain or URL into the context window as verifiable source provenance during RAG extraction?
The 5D Prompt Matrix & Intent Architecture
A metric is only as good as the portfolio behind it. Every prompt in your sampling pool must be classified across five operational dimensions:
5 Portfolio Dimensions
- Product: The dedicated software product, tool, or tier.
- Feature: The specific operational capability or integration.
- Solution: The business use case or challenge being solved.
- Industry: The vertical market or regulatory domain.
- ICP: The practitioner or decision-maker asking the question.
- Brand Status: Branded (explicit name query) vs. Unbranded (brand-agnostic category sweep).
4 Generative Intent Stages
| Intent | Query Focus | Measurement Objective |
|---|---|---|
| Discover | How-to / Problem Framing | Does the model introduce your entity when framing solutions? |
| Research | Best-of / Landscaping | Is your brand recognized as an established category participant? |
| Compare | Vs. / Bake-offs | Are your trade-offs, features, and pricing accurately positioned? |
| Validate | Reviews / Proof Points | Does external consensus hold up when evaluating credibility? |
Governance: Control vs. Discovery Portfolios
If you swap 30% of your prompt set between quarters and your Mention Rate jumps 10 points, did your visibility improve, or did you just change the ruler?
The Control Portfolio
A fixed, strictly versioned prompt set used exclusively for longitudinal benchmarking. Never altered mid-flight; updates are documented as version milestones.
The Discovery Portfolio
A fluid, experimental prompt pool capturing emerging customer language, shifted buyer prompts, and competitor releases. Feeds future control sets.
Companion Tool
Mention Rate Tool (Beta v2.0)
Test prompt portfolios, calculate true presence rates across multi-run distributions, and track statistical stability without relying on single-screenshot anecdotes.