How the score works

What we measure

Every scan asks 15 buyer-intent questions — the kind of question someone deciding between products would actually type — across ChatGPT, Perplexity, and Google AI Overviews. We store every answer verbatim. Everything below is computed from those stored answers.

Mention tiers

Each stored answer is scored against the highest tier it qualifies for:

When more than one tier applies to an answer, the highest one counts.

Position matters

Being mentioned isn't the whole story — where you land among tracked brands matters too. The first brand mentioned scores ×1.0, the 2nd or 3rd ×0.8, and 4th or lower ×0.6. Brands we aren't tracking never push you down a position — only tracked competitors count.

Per-engine score

Your score for a given engine is the mean of tier × position across that engine's valid answers, scaled to 100. An engine needs at least 8 valid answers out of 15 to count toward your overall score at all. For Google, “valid” means an AI Overview actually appeared for that prompt — we report how often that happens separately, rather than treating a missing Overview as a zero.

Overall score

Your overall score is a weighted mean across qualifying engines: ChatGPT 0.40, Google AI Overviews 0.35, Perplexity 0.25 — weighted by roughly how much buyer traffic each surface carries. When an engine doesn't qualify, we drop it and renormalize the remaining weights rather than scoring it as zero.

Share of voice, visibility %, citation share %

• Share of voice — your share of total mention points across all tracked brands.
• Visibility % — the percent of valid answers that mention your brand at all.
• Citation share % — your percent of all citations in the tracked set.

Why trends stay honest

Prompt sets are frozen per version, and the scoring formula itself is versioned — a change to how scoring works shows up as a version change, not a silent jump in your history. If which engines qualify for your overall score changes, we mark it as a baseline reset on your chart rather than blending it into the trend line. Engine outages appear as gaps on your chart, never as score losses, and alerts never fire on missing data.

Limits, stated plainly

15 prompts per scan is a deliberate sample, not a census, and we re-scan weekly rather than continuously. Answers vary run to run — which is exactly why we score across 45 answers, not one.