Settings
Environment
claude-sonnet-4-6Owned Domains
Sources matching these domains receive owned_brand classification and quality tier bonus.
nike.comnews.nike.comabout.nike.comjordan.comconverse.comCompetitor Dictionary
Framing Tags
Scoring Weights
Presence (max 25)
Mention Rank (max 25)
Directness (max 20)
Framing (max 15) · Source Support (max 15)
Framing: tag-based, see tag weights above. Source: owned domain +6, multi-domain +4.
Recommendation Action Types
How Scoring Works
Query Visibility Score (0–100) is a composite score with five transparent components: Presence (25), Mention Rank (25), Directness (20), Framing (15), and Source Support (15). All weights are defined in config/index.ts.
Weighted Visibility Index (WVI) is the priority-weighted average of the latest Query Visibility Score across all active queries: Σ(score × priorityWeight) / Σ(priorityWeight). Queries with priorityWeight=5 count five times more than weight=1.
Framing Score is the sum of per-tag score impacts, clamped to 0–15. Positive tags (premium, technical, trusted) add points. Mixed tags (expensive, legacy) are neutral or slightly negative. Negative tags (overrated, eco_concern) reduce the score.
Source Support Score rewards owned domain citation (+6), breadth of distinct credible domains (+4 for 3+), and penalises zero-source answers.
This score is a directional brand intelligence tool, not a scientific measurement. Use it to identify patterns, prioritise action, and track relative change over time, not as an absolute truth.
How Recommendations Work
Query-level recommendations are generated immediately after each run using deterministic rules applied to the analysis output. Each rule checks specific conditions (e.g. brand absent AND priority ≥ 4) and produces a typed recommendation with an evidence payload that links back to the specific run.
Cluster recommendations are generated across the latest run for each active query. The engine detects patterns such as repeated absence across a category, or eco_concern framing appearing multiple times in skepticism queries.
Global recommendations are generated from aggregate statistics across all latest runs. They fire when thresholds are exceeded (e.g. absence rate > 40%, eco concern rate > 30%).
All recommendation rules are implemented as readable application logic in src/lib/recommendations/engine.ts. No ML or opaque AI reasoning is used. Recommendations are guidance informed by evidence, not certainty.