The five numbers an AI visibility dashboard reports (named, named first, cited, sentiment, sources), what each one actually measures, how Rightcited counts them, and the three ways a number can mislead.
AI visibility is the measurable part of a plain question: when a buyer asks an AI assistant about your category, does the answer name you, and is what it says true? Every tool and every service in this market reports some version of five numbers. This page says what each one measures, how we count it, and where each one can mislead.
The number of buying answers that name your brand as an option, out of the answers checked: offered in 5 of 15. It is the base rate of being in the room. We count a brand as named only when the answer offers it for the buyer's ask; a brand that appears in a sentence like "unlike X, which does not do this" is not offered, and a brand the assistant read a page from without naming is "read but not named", which is its own row because it means the assistant opened your page and found nothing to quote.
The number of answers where your brand is the first option offered for the buyer's ask. First means the first firm offered, not the first name that appears in the text: an answer that opens "unlike Acme, these three are worth a look" names Acme first and offers it never. Most tools call this position or rank; the useful version is the count of firsts with a denominator, next to the same count for the rivals the answers name.
The number of answers that read a page on your domain, out of all answers checked, including the brand questions. Citation is not the same as being named: an assistant can read your page and quote it for a rival (we have watched a resort group's own listicle cited for a competitor's villas), and it can name you from a directory without ever opening your site. The pair of numbers, named and cited, tells you which problem you have: a site the assistants do not read, or a site they read and do not quote.
Asked directly whether a company is legitimate, or what reviews say, an assistant gives a verdict. We grade each one by hand into four: favorable, favorable with a caveat (with the caveat quoted and traced to the page it came from), negative, and wrong company (the assistant answered about a different entity with a similar name). Wrong company is the strongest evidence a check can produce and goes in the headline. A tool that scores sentiment from minus one to one hides which of these happened.
For every answer, which domains the assistant read, and who owns each: you, a rival, a community thread, or an independent site. This is the number that becomes a work plan. When four of the six pages behind a category answer are rival brands' own roundups, the fix is a page and an inclusion request; when the answer is written from a directory entry with no city, the fix is a correction; when it is written from your own blog post that lists sixteen rivals, the fix is on your site.
A share-of-voice percentage with no denominator. Visible in 62% of answers means nothing without the number of answers, the questions asked, and whether the assistant searched. Ask for the count and the list.
A sentiment score. A brand answer that says "legitimate, but I'd book the property, not the company" is favorable with a caveat, and the caveat is the whole finding. A score of 0.7 loses it.
An absolute citation count from an index. An index of archived AI answers is a flight recorder: useful for change over time and for the gap to competitors on the same panel, and unreliable as an absolute, because it undersamples niche and local categories. We report index counts only against the same index earlier, and we say which index basis we used.
A number is only worth collecting if a decision follows from it. Named and named first tell you where to build; cited tells you whether the site or the sources are the problem; sentiment tells you which records to correct; sources tell you whose pages to request, correct or displace. A dashboard that reports the five without the answers behind them is a report of a problem; the answers are the work plan.
Share of voice is one presentation of the first number (named) as a percentage across a prompt set. It is useful next to the same figure for rivals on the same prompts, and misleading on its own, because it hides the denominator and the questions.
Enough to cover the questions buyers in your category actually ask, which for one brand is usually about fifty: category, product, comparison, trust and alternatives questions, in the buyer's words. More prompts cost more to read, and unread prompts change nothing.
Because a verdict should change when reality changes, and the assistants answer with some randomness. Weekly checks with a two-check rule give a count that moves when a page changed, not when a model drew a different page on a Tuesday.
Or start smaller: the AI Page Check runs one page through all four assistants for $2.99.