How AI Visibility Scores are measured

What we measure

We ask AI assistants the kinds of questions real buyers ask — "what are the best [category] tools?", "[product] vs [competitor]", "best alternatives to [product]" — and record whether an entity is mentioned, where it ranks among the entities named, and the sentiment of the surrounding text.

Every public report shows the prompts used and excerpts of the actual AI answers, so you can verify any number on the page against its source.

The score

AI Visibility Score = 50% mention rate + 30% average rank (normalized over positions 1–10) + 20% sentiment. It is a comparative signal for one prompt set at one point in time — not an absolute market-share measurement.

When a report covers multiple AI providers (Pro), the headline score is the average of per-provider scores.

Models and runs

Free public reports track an OpenAI GPT-4-class model. Pro reports additionally compare Anthropic (Claude) and Google (Gemini) models where enabled.

AI answers vary between runs. Claimed reports run each prompt multiple times (currently 3) and aggregate: an entity counts as mentioned when a majority of runs mention it, rank is the median across runs, and the report shows a recommendation-consistency percentage. Unclaimed directory reports are single-run and should be read as a snapshot.

Update cadence

Public reports refresh on a weekly rescan cycle, claimed reports first. Week-over-week movement feeds the trending and movers pages.

Known limitations

Mention detection is exact-name-based and can miss aliases, abbreviations, or indirect references ("the design tool from Figma's biggest rival").

Sentiment on free reports uses keyword analysis near the mention, which can misread sarcasm or complex phrasing. Pro reports add an LLM verification pass that also extracts positioning themes and content gaps.

Prompt sets are finite. A score reflects the tested prompts — not every possible buyer question. Claimed owners can edit prompts to match the questions their buyers actually ask.

AI assistants change model versions without notice; scores are comparable within a scan and directional across weeks.