Read the numbers behind AI visibility.

Mention rate starts with assessed answers.

Brand mention rate is the number of saved answers that mention a brand, divided by the number with a recorded assessment of that brand. Multiply by 100 to express it as a percentage. An answer awaiting analysis is outside that denominator.

For example, suppose you saved 20 answers. Sixteen have an assessment of your brand, and eight of those mention it. The mention rate is 8 ÷ 16 × 100 = 50%. Four answers are still unassessed. Calling the rate 40% would wrongly treat those four as confirmed absences.

Check the denominator for each competitor too. A brand assessed in 10 answers and another assessed in 20 answers do not have identical coverage, even if the percentages look directly comparable.

Average rank describes a smaller subset.

Recommendation rank records an explicit recommendation position when the answer supports one. An incidental mention or an unordered recommendation can have no explicit rank. Missing rank does not assign a product the last position.

If the recorded positions are 1, 2, and 3, the average rank is 2.0. Two additional mentions without positions do not enter that average. Read this as an average over three ranked answers, not over all five mentions.

Lower positions mean earlier recommendations within the answers that were ranked. A lower average can still coincide with fewer mentions overall, so read rank and mention rate together.

Coverage tells you what was observed.

Questions sampled counts distinct questions with at least one saved answer in the selected period. Asking the same question five times does not create five different sampled questions. Platforms sampled counts providers with a saved answer, rather than every enabled provider.

For example, 10 questions across three providers creates 30 planned question-provider combinations for one collection round. If only ChatGPT has completed those questions, the question count may be 10 while the other two providers still have no saved results. Open Tracking to inspect the provider-by-question coverage.

Queued, running, failed, saved, and awaiting assessment describe different stages. A failed collection is missing data; it is not evidence that a provider omitted your brand.

Separate percentage points from percentage change.

A mention rate moving from 40% to 50% increased by 10 percentage points, often abbreviated pp. Relative to the original rate, that is a 25% increase. Those are different calculations describing the same movement.

Before comparing periods, check the question set and provider mix. Adding questions where your brand is already common can raise the aggregate rate without changing any existing answer. Repeated captures of one question can also carry more weight in an answer-based rate.

Treat a comparison based on very few answers as a lead to investigate. The dashboard summarizes the answers you collected; it does not estimate what every buyer sees or establish that a content edit caused a change.

Keep citations and collection timing separate.

A citation records a source reference. It does not automatically mean a brand was recommended. In Echuno, API tracking has a separate report: its website citation rate uses completed API answers as the denominator, while its brand visibility rate uses answers with recorded brand analysis. Check which report you are reading before combining figures.

Collection duration measures the time from starting a request to saving its answer. Queue-to-save duration also includes waiting for the browser runner, other work, and maintenance pauses. A seven-minute total with a 30-second collection does not mean the provider spent seven minutes generating the answer.

Finish by opening the evidence.

In the overview, open the mention-rate card to inspect assessed answers. Use the rank card to read responses with recorded positions. Filter the saved Answers library by provider and question to check whether the summary matches the wording you actually collected.

For a team update, record the period, question set, providers, assessed-answer count, and ranked-answer count beside the headline figures. Then choose one or two supporting answers to read in full. A clear denominator and an exact quote are more useful than an unexplained score.

• Scope: which dates, questions, and providers were included?
• Coverage: how many answers were saved, assessed, and explicitly ranked?
• Evidence: what does a representative saved answer actually say?
• Follow-up: what specific observation should the next collection revisit?