How is the AI visibility score calculated?

Score = 100 × (0.60 × mention rate + 0.25 × citation rate + 0.15 × prominence), where prominence is the average of 1 divided by the brand's position among the category's brands, counting 0 for answers that do not name it.

How many answers is each leaderboard based on?

Each prompt is sent once to each of 3 engines per run, so a category's sample is its number of prompts times the engines that answered. The latest runs have between 47 and 60 answers per category.

Why does a brand move between weeks?

AI engines do not return the same answer every time, so part of any week-to-week change is sampling noise. Treat moves of a few points, or one or two ranks, as within noise.

AI visibility leaderboards

Methodology

What is measured, how the score is computed, and how much a single week's ranking can be trusted.

Collection

Each category has a fixed list of brands and 17 to 20 buyer-intent prompts, such as "best rank tracking tool for agencies". Every run sends each prompt once to ChatGPT, Gemini and Google AI Mode as an async task through this API, with the country set to US. The prompts are published on each leaderboard page exactly as sent. A task that fails is left out of the sample; the page names the engines that returned no answers.

What counts

Mention: the brand's name or a listed alias appears in the answer text as a whole word, ignoring case. Brand names that are also everyday words (for example Heap or Mention) are matched case-sensitively.
Citation: one of the answer's sources or inline citations links to the brand's domain or a subdomain of it.
Position: 1 plus the number of the category's brands whose first mention comes earlier in the same answer. Brands outside the list do not affect it.

Score

score = 100 × (0.60 × mention rate + 0.25 × citation rate + 0.15 × prominence)

mention rate  = answers naming the brand / answers
citation rate = answers citing the brand's domain / answers
prominence    = sum over answers of 1 / position (0 when not named) / answers

All rates pool the answers of every engine that responded. A brand named first in every answer and cited in every answer scores 100. Movement is the change in rank against the previous run of the same category. Each snapshot keeps the raw per-engine counts, so any rate can be recomputed.

Limits

Sampling variance. Engines answer the same prompt differently from one request to the next, and each prompt is asked once per engine per run. With the smallest current sample of 47 answers, a mention rate near 50% has a 95% margin of roughly ±14 percentage points. Small moves between weeks are usually noise.
The prompts shape the result. A different set of questions, wording or country would rank brands differently.
Text matching is literal. Misspellings and product names that are not listed as aliases are missed; a name used in another sense can be counted.
Engine mix. When an engine fails for part of a run, the pooled rates lean toward the engines that answered.
Only listed brands are ranked. A brand the answers recommend but the category does not list is not counted.
The ranking measures visibility in AI answers, not product quality.

All leaderboards · Track your own prompts