# AI Visibility — 10 Metrics

> Complete list of all metrics and queries in the Be1st.ai AI Visibility tool.

6 queries total | 3 AI models | 18 evaluations total

> Requires Web Fingerprint — uses enrichment data and embeddings for hallucination detection.

## Knowledge queries — direct questions about the domain

*3 direct queries × 3 AI models = 9 evaluations*

### K1 — Identity and profile

Does AI know your domain? Who is behind it, what industry it operates in, main services/products, target audience, strengths and weaknesses.

> *“Do you know the website {domain}? Create a profile…”*

### K2 — Competition

Who does AI consider your competitors? Automatic extraction of competing domains.

> *“What are the main alternatives or competitors to {domain}?”*

### K3 — Recommendation

Would AI recommend your website? Measures the willingness of models to actively recommend.

> *“Would you recommend the website {domain} to someone?”*

## Discovery queries — indirect questions without mentioning the domain

*3 AI-generated queries × 3 models = 9 evaluations*

### D1 — Main topic

AI generates a question about your main topic without mentioning the domain — tests spontaneous recommendation.

> *“I'm looking for [main topic of the website]. What websites would you recommend?”*

### D2 — Category

Question about your category/segment — does your website appear among the recommendations?

> *“What are the best [category] solutions?”*

### D3 — Keyword combination

Question combining your keywords — measures organic visibility.

> *“I need [keyword 1] and [keyword 2]. What platforms do you know?”*

---

## Knowledge Levels

| Level | Score |
|---|---|
| none | 0 pts — AI does not know the domain at all |
| confused | 10 pts — knows it but gets it wrong (hallucination) |
| partial | 40 pts — partial knowledge |
| good | 75 pts — good knowledge, correct profile |
| excellent | 100 pts — excellent, complete profile |

## Data collected from each response

- Company profile — type, industry, services, target audience, region
- Reputation — sentiment (positive/neutral/negative), strengths and weaknesses
- Competitors — automatically extracted domains with mention frequency
- Ranking position — at which position AI recommends your domain (1-5+)
- Direct link — whether the answer contains a direct HTTP link to the website
- Model confidence — how confident the model is (0.0-1.0)

## Hallucination Detection

Each response is compared against the website fingerprint using cosine similarity of embeddings.

- Similarity ≥ 0.58 — response consistent, no score change
- Similarity 0.50-0.57 — partial match, knowledge level capped at "partial"
- Similarity < 0.50 — hallucination! Downgraded to "confused"

## Scoring

Base score = knowledge level (none=0, confused=10, partial=40, good=75, excellent=100)

Bonuses: +5 direct link, +5 confidence ≥ 0.8, +5 positive sentiment

Penalty: -10 response processing failed (fallback mode)

**Overall visibility score = average of all individual scores (0-100).**

## Tested AI Models

- The current list of tested models is managed in Pipeline Config (admin) — typically 3–4 models in parallel (e.g. OpenAI GPT, Google Gemini, Perplexity Sonar).