AIVOLabs
Methodology

How the AI Visibility Score is calculated.

Scoring is explainable and versioned. This page documents prompt types, detection rules, and weighted signal families.

AI Visibility Score0out of 100
The score

A single number, 0–100.

The AI Visibility Score is a weighted average of three signal families. Brand-neutral signals drive the majority of the score, while brand-recall remains a smaller context signal.

score =0.65·Organic Visibility + 0.25·Competitive Presence + 0.10·Brand Recognition

Signal families

Organic Visibility

65% weight

Primary signal from brand-neutral discovery prompts. Measures whether your brand appears when buyers ask category questions without naming you.

weighted_mentions_from_discovery_blind / discovery_blind_tests

Competitive Presence

25% weight

Secondary signal from brand-neutral comparative prompts. Captures whether you are present and how strongly you rank against alternatives.

weighted_visibility_from_competitive_blind / competitive_blind_tests

Brand Recognition

10% weight

Tertiary informational signal from explicit brand-recall prompts. Included for context only and intentionally capped so recall cannot dominate score.

weighted_visibility_from_brand_recall / brand_recall_tests

The scan pipeline

01

Generate prompt set by type

The system generates a fixed prompt set with explicit prompt types. Discovery and competitive prompts are brand-neutral by default.

02

Fan out to AI platforms

Each prompt runs across ChatGPT and Gemini. Results are stored per prompt-platform test with deterministic parsing metadata.

03

Parse responses for mentions

Parser-side detection uses normalized domain, display brand, and deterministic token variants to classify mention signal and rank position.

04

Compute the score

Visibility score is weighted by signal family: Organic Visibility (0.65), Competitive Presence (0.25), and Brand Recognition (0.10).

Assumptions & limits

  • Blind prompts are guarded. If a brand-neutral prompt contains your brand/domain token, generation fails fast.
  • Methodology is versioned. Runs carry prompt set and scoring versions so historical comparisons remain traceable.
  • Detection is deterministic. Mention parsing uses substring/list rules with rank extraction and no fuzzy matching.
  • Legacy runs may differ. Older runs can be marked as legacy if they include branded prompt leakage.