How the numbers are made

A data product is only as good as its honesty. This page explains exactly what we measure, how, and where the limits are.

1. Buyer prompts

For each scan we build ~12 prompts from your brand, category and competitors — the questions real buyers ask assistants:

• “What are the best {category} tools right now?”
• “What are good {brand} alternatives?”
• “{brand} vs {competitor} — which one is better?”
• “Cheapest reliable {category} that is still good?”
• …and similar best-of, review and use-case phrasings.

2. Engines

Prompts run against AI answer engines — currently Perplexity (with real citations) and Google AI Overviews, with more being added. We record each answer’s text and the sources it cites.

3. What we extract

• Mentions — is your brand named at all, and in what order relative to competitors;
• Citations — which domains the engine used to build the answer;
• Displacement — prompts where competitors are named and you are not.

4. The Visibility Score (0–100)

Three weighted components:

• 50% mention rate — share of prompts where your brand appears;
• 30% position quality — how early you appear among mentioned brands (1st = 100%, 2nd = 70%, 3rd = 50%, later = 30%);
• 20% own-domain citations — share of cited answers that cite your own site. When an engine returns no citation data, weights renormalise to 62.5/37.5.

Share of voice is your brand’s share of all brand mentions across the scan’s answers.

5. Monitoring & alerts

Tracked brands are rescanned daily. We diff consecutive scans per engine+prompt pair: newly lost prompts (you disappeared — and who took the spot), recovered prompts, and share-of-voice shifts. A displacement or a score drop of 8+ points triggers an email alert.

6. Honest limitations

• AI answers are probabilistic. The same prompt can produce different answers between runs. Single scans are directional snapshots; trends across daily scans are the reliable signal.
• Leaderboards are point-in-time. Category pages state their scan date and engine; they are not a rolling average.
• Coverage is partial. We query specific engines with specific prompts — not every phrasing a buyer might use.
• No gaming involved. We only read answers; we don’t (and can’t) influence what engines say.

Questions?

[email protected] — we’re happy to defend every number.