Metrics defined

Every RankX AI metric is a ratio with a stated denominator, and the denominator is the part that decides what the number means. Mention rate divides by checks with a known verdict. Share of voice divides by the same. Shopping visibility divides by answers that showed products at all. A metric with no denominator available is reported as unknown, never as zero.

This page defines each metric once, so the rest of the documentation can refer to it rather than redefining it.

Known checks and total checks

These two are not a metric; they are the reason the metrics are trustworthy, and they appear beside almost everything.

• Total checks is every check that completed, analysed or not.
• Known checks is every check that produced a determined verdict.

Known checks is the only honest denominator, and it is the one RankX AI uses. A check that ran but could not be analysed contributes to neither the numerator nor the denominator, and it is counted separately so the gap stays visible.

If known checks is zero, every rate built on it is unknown. That is why a fresh Website shows blanks rather than zeros.

Mention rate

What it is. The share of known checks in which your brand was named, reported per assistant over a 7, 30 or 90 day window.

How it is calculated. Mentioned checks divided by known checks, per assistant.

What it does not mean. It is not a share against competitors, and it is not comparable across assistants as a single number: the five assistants disagree with each other by design, and a 40% mention rate on one and 10% on another is two findings rather than an average of 25%.

When it is null. No check in the window has an analysed verdict. Report it as unknown. Reporting it as 0% tells a customer they are invisible when the truth is that nothing has been measured yet.

Share of voice

What it is. How often your brand is named in AI answers compared with the rivals named in those same answers. It is the platform's headline visibility figure, and it is deliberately comparative.

How it is calculated. Your mentions divided by known checks, alongside the same figure for each rival brand found in the answers. Rivals are folded by brand, so different spellings of one company count once.

Why comparative. Because a new business is genuinely absent from AI answers, which is usually why they signed up, and a bare "0%" is both demoralising and uninformative. "This rival appears in 21 of 25 answers and you appear in none" is the same fact, and it is actionable.

Two zeros that mean opposite things. Share of voice of null means no verdict has been produced yet. Share of voice of 0 means RankX AI measured and your brand was genuinely absent from every analysed answer. The first is a gap in the sample, the second is the finding worth acting on, and RankX AI never renders them the same way.

Related figures shown beside it. Your rank among the brands compared, how many brands were compared, how many distinct assistants named each rival (breadth rather than volume), and the single most-named rival.

Position in an answer

What it is. Where your brand sat in an answer that contained a ranked list.

When it exists. Only when the answer actually ranked things. Many AI answers are prose, and prose has no position. A null position means the answer had no ranking, not that you ranked last.

Google position

What it is. Your position in ordinary Google results for a tracked keyword, checked to depth 20.

The three states, which are never merged.

Collapsing the second and third into "no position" would turn an unmeasured keyword into a failing one. RankX AI also reports why a scheduled check was skipped, when one was.

AI Overview presence

What it is. How often Google's AI Overview appeared for your tracked keywords, and whether your brand was named or cited inside it.

A real third state. "Unknown" means an overview was detected but not captured. It is not "you were not mentioned". Treating it as a negative would manufacture absences that were never measured.

Where it comes from. The rank tracker, on tracked keywords, alongside the ordinary position check. It is not part of the scheduled assistant run. See citations, two kinds.

Shopping visibility

What it is. Whether your products are recommended when a shopper asks an AI assistant, per surface, over the last 28 days.

How it is calculated. Answers containing your products divided by answers that showed products at all. An answer that returned no product carousel has no opinion about your catalogue, so counting it against you would be inventing a loss.

When it is null. Nothing showed a carousel in the window, so there is no denominator. Never rendered as zero.

Reported alongside it. Checks run, checks that produced an answer, checks that showed products, checks that could not run, and how many answers included your products, per surface. Six states, and only one of them means "absent".

AI Readiness score

What it is. A 0 to 100 score for whether AI assistants can reach, read and understand your site.

Two rules that make it unusual, and both are deliberate.

First, the overall score is the weakest category, not the average. Averaging made a site with a broken machine-readable layer score 85, because table-stakes passes padded the denominator and the one real defect diluted away. The weakest category is the more useful number because it names the thing to fix, and the score always renders beside that name.

Second, below a coverage floor there is no score at all. Not 0, which reads as "your site is broken", and not 100, which reads as "nothing to do". A score never renders without its coverage fraction beside it, and a check that could not run contributes nothing in either direction.

The checks the product actually runs, with the category and weight of each, are in the AI Readiness check reference, which is generated from the product rather than typed.

Website Audit score

What it is. A 0 to 100 score for the technical health of the crawled site.

How the penalty is weighted. Each issue carries a weight, and the penalty is a ratio of the right denominator for that issue class: pages crawled for a per-page issue, the sitemap for a sitemap issue, and the site itself for a site-wide one. A missing sitemap is one site-wide fact and is not diluted by the number of pages crawled.

Not assessed is not passed. A check whose audit path did not run is reported as not assessed. See what "a check did not run" means.

Content quality and GEO scores

What they are. Two scores on a generated article: a content quality score, and a GEO score for whether the piece is structured to be extracted and cited.

When they are null. Not yet computed. A null score on a draft is a schedule, not a zero.

Metrics RankX AI does not report

Worth stating plainly, because competitors report them and their absence here is a decision rather than an oversight:

• Sentiment. RankX AI does not report brand sentiment in AI answers. There is no sentiment figure on any screen and no sentiment field in any response. A sentiment column was designed and removed before it shipped rather than being filled with a guess.
• A single blended "AI visibility score" across all five assistants. The per-assistant figures are the real figures; a blended headline would hide the disagreement that is the most useful thing in the data.
• Estimated traffic from AI answers. Nobody can measure this honestly. Where RankX AI reports traffic it comes from Search Console and Analytics, which are first-party and named as such.

Where to go next

• Null is not zero, for the rule every metric on this page obeys.
• How RankX AI measures visibility, for where the checks come from.