Technical Documentation

Weighted Visibility Score (WVS) — Methodology

The complete rubric LLM Check uses to measure how prominently and accurately a brand is cited by ChatGPT, Gemini, and Perplexity.

Overview

The Weighted Visibility Score (WVS) quantifies a brand's "Share of Voice" in AI-generated responses. Each scan fires 5 intent-based prompts against 3 LLMs (ChatGPT, Gemini, Perplexity), producing 15 scored data points per scan. Every data point is evaluated across four dimensions, each contributing a defined number of points to a 0–100 score.

20 pts

Mention

30 pts

Prominence

30 pts

Sentiment

20 pts

Citation

Scoring Dimensions

A

Mention — 20 points

A binary gate. The brand's root domain, brand name, or known alias must appear in the LLM's response for any other score to apply. If there is no mention, the score for that data point is 0. Detection checks root domain URL, exact brand name (word boundary), and registered aliases.

ResultPoints
Brand found in response20
Not found0 (scan ends here)
B

Prominence — up to 30 points

Where in the response the brand appears. Being in the opening paragraph of a ChatGPT answer signals primary recommendation. Appearing only in a long "alternatives" list at the bottom signals low authority.

LevelConditionPoints
HighBrand appears within first 150 words of response30
MediumBody text, small list (2–5 items), general recommendation15
LowAfter "See Also" / "Further Reading" markers, or list of 10+ items3
C

Sentiment — up to 30 points

LLM-as-judge scoring. GPT-4o-mini evaluates the surrounding context of the brand mention and rates the strength of the recommendation on a scale of −1 to +1. The value is linearly mapped to 0–30 points. A purely neutral mention ("Brand X is a tool for Y") scores 0.5 on the scale (15 pts). An explicit top recommendation scores close to 1.0 (30 pts). A warning against the brand scores close to −1 (0 pts).

Sentiment valueMeaningPoints
+1.0Explicitly recommended as top choice30
0.0Neutral / purely factual mention15
−1.0Explicitly warned against0
D

Citation (Provenance) — 20 points

A direct, clickable URL pointing to the brand's root domain must appear in the LLM response. This is the highest-trust signal — it means the AI model is actively directing users to the source, not just mentioning the brand by name. Detected via URL pattern matching against the brand's root domain.

ResultPoints
Direct URL to brand domain present in response20
No URL (name-only mention)0

Query Methodology

LLM Check generates 5 intent-based prompts per scan using GPT-4o, tailored to the site's industry and content. Each prompt is sent to all three models in parallel, producing 15 independent responses. Prompts are constructed to reflect realistic user queries ("What is the best tool for X?") and include a suffix instructing the model to name specific tools and companies, maximising the chance of brand-level responses.

ChatGPT

GPT-4o via OpenAI API

Gemini

gemini-1.5-flash via Google AI

Perplexity

sonar model via Perplexity API

See your WVS score

Run a free LLM Readiness audit, then unlock the full AI Visibility scan to get your Weighted Visibility Score.