How we measure AI visibility
Our methodology is fully transparent. Every point in your score traces back to a specific prompt, a specific AI response, and a specific result. No black box.
Where supported, we normalize each prompt-model or prompt-surface result across 3 sampled responses in the same run. That helps reduce one-off answer swings and lets us show confidence, not just a single snapshot.
See the full version history: Methodology changelog →
The AI Visibility Score
The AI Visibility Score is a weighted number from 0 to 1000, modeled after a credit score. Higher is better. A score of 0 means your business did not appear in any AI-generated answer across the prompts we tested. A score of 1000 would mean your business appeared first, prominently, and without competitors in every answer.
Mention detection
We check whether your business name appears in each AI response. This is the baseline — you either showed up or you didn't.
Rank position
AI models often list multiple businesses. We detect where in the answer your business appears — first mention scores higher than fifth.
Sentiment
We analyze the language used when your business is mentioned. Positive framing scores higher than neutral; neutral scores higher than negative.
Competitor presence
If your business is mentioned alongside many competitors, that dilutes visibility. Being the only or primary recommendation scores higher.
How prompts work
We run prompts that simulate real questions your potential customers ask AI models. Examples include "What are the best options for X?" or "Which companies do you recommend for Y?" We run these prompts across multiple large language models and collect the responses.
Each prompt is run independently across all monitored models or consumer surfaces. When sample-based normalization is enabled, each target is run 3 times inside the same snapshot. We then aggregate those repeated responses into one normalized prompt result before rolling up category and overall scores.
Confidence and stability
AI responses are non-deterministic. The same question can produce different answers a few seconds apart. Repeated sampling lets us estimate whether a result is stable or noisy instead of treating a single answer like ground truth.
High confidence
Repeated samples largely agree. These results are the strongest signal of true visibility.
Mixed confidence
Some repeated samples agree and some diverge. Visibility is present, but less consistent.
Volatile
Repeated samples disagree materially or fail often. These results are still shown, but they are discounted slightly in scoring.
Which AI models we monitor
We currently monitor:
- → ChatGPT — GPT-4o Mini and GPT-5.2
- → Claude — Opus 4.6 and Sonnet 4.5
- → Gemini — 2.5 Flash
We report scores per model and in aggregate. Visibility varies significantly across models — a business can be consistently recommended by one model and absent from another.
Score ranges
| Score | Label | What it means |
|---|---|---|
| 800–1000 | Dominant | Consistently first or primary recommendation |
| 600–799 | Strong | Frequently mentioned, competitive position |
| 400–599 | Moderate | Appears in some answers, inconsistent coverage |
| 200–399 | Weak | Rarely appears, significant gaps |
| 0–199 | Not visible | Little to no presence in AI-generated answers |
Transparency commitment
We believe AI visibility data should be as transparent as possible. That means:
- →Every score component is labeled and traceable to source data
- →Methodology changes are versioned and published in the changelog
- →Raw AI responses are stored and available to review
- →Normalized runs retain the underlying sampled responses for auditability
- →Score recalculations are supported when methodology updates
See your score
Run a free scan to see exactly how your business is measured — prompt by prompt, model by model.
Run free scan