What a run is
A run is one tracked prompt asked of one engine once. We collect the answer through the engine’s official API with web search switched on, store the full text and any sources it cited, then a model reads the answer and records, for your brand and each tracked competitor, whether it was mentioned, roughly in what order, and with what sentiment. A mention counts when the answer uses the name, or any alias or abbreviation you have set (or one derived from the name, such as “BCA” for “Bank Central Asia”).
Answers from an API can differ from what a person sees in the consumer app, and the same prompt can give different answers on different runs. Treat single runs as samples and look at trends.
Visibility score
For one engine: the runs that mentioned your brand divided by the runs that completed, on that engine’s most recent day with runs, as a percentage. The headline score is the average of those per-engine scoresacross the engines that have data, so an engine with few runs counts as much as one with many. It is a single-day snapshot, labelled “as of” its date, and it is kept until a newer day replaces it. The trend charts show the same score per day.
The headline can include engines that are not on your plan’s run list, because it covers every engine with data. The per-engine breakdown beside it shows which engines contributed.
Share of voice
Over the chosen period, every mention of your brand or a tracked competitor is counted, and each name’s share is its count divided by the total, rounded to a whole percent. Your own brand is included and marked. Names with no mentions are left out, so the rounded shares may not add to exactly 100.
Position and sentiment
Position is the order in which the brand first appears among all brands in the answer (1 is first). Sentiment is positive, neutral or negative as judged from how the answer talks about the brand; a plain factual mention is neutral. Both are model judgements, so they are labels to scan, not measurements. The Sentiment page counts them per day, gives a net score (positive minus negative, as a share of mentions), and can group the wording into themes. Theme grouping is done by a model; the counts are ours.
Citations
A citation is a web page an engine listed as a source for an answer. Repeats of the same page are collapsed. Citation insights compares each source this period with the period before: a source is fading if it had at least 3citations before and is now cited half as often or less. “First cited” is the first time we recorded that page, which can be later than when an engine first used it.
Query fan-out
The web searches an engine ran behind the scenes to answer a prompt. We read them from the stored answers, so only engines that report them appear: Gemini and Claude do; ChatGPT’s search and Perplexity do not, and are shown as “not reported” rather than as zero.
Topic matrix
A topic is a prompt’s tag. For each, visible is the share of its answers that mention you, and cited is the share that cite a page on your own domain. A topic with 50% or more on a measure counts as high on it, which gives four groups: strongholds, visible but not cited, cited but not visible, and needs attention. Topics with fewer than 3 answers are listed but not placed.
Demand research
Estimated relative demand from AI research, not measured search or prompt volume. Use it to compare questions against each other, not as absolute numbers. It does not come from a measured data source: a model, using a live web search when one is available, lists the questions people plausibly ask and scores them against each other.
AI crawler analytics
Counts of visits to your site by known AI crawlers, from the requests or logs you send us. A user agent can be faked and some assistants fetch pages with an ordinary search-engine crawler that cannot be told apart, so counts are a strong signal rather than proof.