How the Visibility Score is calculated.
Every score ships with the prompt set that produced it, the raw runs behind it and a confidence interval — so you can check our work instead of trusting it.
1. Build the prompt corpus
Between 25 and 250 questions your ICP actually asks, drafted from your strategy docs by the AI Visibility agent and editable by you before the first run — the exact count depends on scan depth (quick, standard or full). Generic category questions ("best CRM for small teams") sit alongside your own brand and competitor names.
2. Run it across engines
Each prompt runs against four engines — ChatGPT, Gemini, Claude and Perplexity — the consumer AI surfaces buyers use to ask category questions today. Brand-name, product-name, comparison (“X vs Y”) and category (“best X for Y”) prompts are sampled multiple times per engine; the long tail runs once. We rerun the high-value prompts because answers to the same prompt vary between calls — a single run is a coin flip, not a score.
| Engine | Surface | Sampling |
|---|---|---|
| ChatGPT | Web browsing mode | 3 runs for brand/product/comparison prompts · 1 run for long-tail prompts |
| Gemini | Web grounding on | 3 runs for brand/product/comparison prompts · 1 run for long-tail prompts |
| Claude | Web-enabled mode | 3 runs for brand/product/comparison prompts · 1 run for long-tail prompts |
| Perplexity | Default search | 3 runs for brand/product/comparison prompts · 1 run for long-tail prompts |
3. Score with an interval
For each prompt we record whether your brand is mentioned, its position in the answer and whether a page of yours is cited. The Visibility Score is the Wilson confidence interval on the mention rate across all runs — it narrows as the corpus and run count grow, and widens honestly when the sample is thin (the free 5-prompt check, for example).
4. Diagnose and act
Every gap — a prompt where you're absent, a competitor cited instead — becomes a task routed to the Content, SEO or AI Visibility agent. The next weekly run tells you whether the fix worked.
Independence
noor grades its own agents against this score. Keeping the prompt set visible and the runs exportable is how you check that grading instead of trusting it.
Questions
Email methodology@heynoor.ai, or export your raw runs from Settings and check the math yourself.