What should a founder do with an AI visibility number before it changes a budget?
An AI visibility number is not a decision until the sentence behind it and the release it came from are both visible. I would not move a budget on a score that blends a cited answer with a string that merely sat near a name.
The useful question is not whether a dashboard went up. It is whether a buyer, asking the question they actually ask, gets an answer built from a source that is not the company, and whether that source says something specific. A number that cannot be opened into those two checks is a label. Labels are easy to report and hard to defend.
I separate three documents that teams keep folding together. The first is the score: a count of mentions, a share of voice, a visibility index. The second is the sentence the score was counted from. The third is the release: which prompts, which engines, which window, and which identifier lets someone else find the same table. If the second or the third is missing, the first does not get to change a plan.
That rule comes from reading the Machine Relations Index as a founder, not from treating the Index as a ranking I am trying to win. The Index is the maintained record. AuthorityTech's measurement page is where the operating metrics are laid out for a team that has to instrument the work. This page is the judgment I apply before I let either of those documents move money.
The judgment has a sequence. Read a sample of the raw answers, not the chart. Split runs that cite a URL from runs that do not. Then ask whether the prompt already contained the brand. A prompt that hands the model the name is a test of whether the detector can avoid scoring the prompt itself. It is not a test of whether a buyer who has never heard of the company would be told the name. Those are different decisions, and a single blended score hides the one that matters for discovery.
I also want the release id on the page that cites the number. A percentage without the release is a number that cannot be checked when the next release moves. The Index changes. A founder decision that cites it has to say which edition it used, or the decision cannot be reversed in public when the evidence changes. I would rather publish a narrower claim tied to one release than a timeless claim that no one can audit.
What I would not do is buy a replacement for the number. Software that reports visibility can be worth its operational job. It is not evidence that the brand will be cited. The evidence is the source the answer uses. If that source is a forum, a video, or a page the company does not control, the work is to earn a place in sources of that kind, not to negotiate with the chart.
Share of voice still has a use. It tells you how much coverage ran. It does not tell you whether an answer engine picked any of it up. I keep the clip report for the humans who read the clips. I do not let it stand in for the citation. When the two disagree, I believe the sentence in the answer, and I go find the page that sentence came from.
The Index number
In that release, the AI Visibility and GEO worth segment had 137 observed answer runs and 467 cited sources. reddit.com was cited in 38 of the 137 runs.
Release mri_score_v2.0+2026-09-18+8fa38e54dd0a, window 2026-05-10 to 2026-09-18. Source.
The definition is Machine Relations, the definition. The execution page is The seven metrics AuthorityTech uses to measure AI search visibility.
Sibling answers
FAQ
Does a higher visibility score mean the brand will be recommended?
Not by itself. A score that cannot be opened into the cited sentence and the release is not evidence of a recommendation. I treat it as a prompt to read the answer, not as the answer.
Which number do I keep if the clip report and the citation disagree?
I keep both, and I let them do different jobs. The clip report says coverage ran. The citation says an answer used a source. The budget follows the second when the buyer's first read is the answer.