A useful score isn't one blended number, it's three separate counts, named, recommended and cited, tracked per engine with the actual answer kept as evidence. A single figure out of 100 hides which of the three actually moved, which is why Reflexa's own check refuses to produce one.
This question usually comes from wanting what the Reflexa agent gives you without installing anything: some sense of where you stand before deciding whether to act. The honest answer is that the standing itself isn't one number, it's a small set of observed counts, and the rest of this guide shows you how to get them.
It depends entirely on which vendor you ask, because the term has no agreed definition. Some tools mean a 0-100 readiness score for how crawlable your site is. Others mean a share-of-voice percentage against named competitors. A few mean how often your brand is mentioned at all, with no distinction from being recommended. None of these is wrong on its own, but comparing a "visibility score" from one tool against another is comparing two different measurements that happen to share a name.
Why it matters: before you chase a number, ask what it is actually counting. If the vendor can't answer that in one sentence, the score is not usable.
Because a single score blends findings that have nothing to do with each other, and the average tells you less than either finding alone. A site with a blocked AI crawler and another with slow, thin pages can land on the identical score while needing completely different fixes. Reflexa's free page check instead runs 35+ structural checks and returns them worst first, each marked Fix, Look, Info or OK in plain words, with the evidence next to it rather than folded into one grade.
Proof: the same reasoning applies to the answer-side measurement. In the Reflexa AI Visibility Index's September 2026 project-management edition, brands named by AI in 27-51% of eligible answers were cited on their own domain in only 6-18% of those same answers, a 21-33 point gap. A single blended score would have reported both brands as roughly "visible" and missed that gap entirely.
Do this instead: read how the page check is ordered and what each verdict means before you look for a single grade anywhere in it. There isn't one, on purpose.
Named, Recommended and Cited, each answering a different question about the same AI answer. Named means your brand appears anywhere in the text, even a passing mention. Recommended is stricter: you have to be one of the top three products the answer actually suggests. Cited is different again, whether the engine lists your own domain among its sources, independent of whether it recommends you at all.
| Count | What it answers | A high count alone means |
|---|---|---|
| Named | Does your brand appear anywhere in the answer? | You're on the radar, possibly only as a runner-up |
| Recommended | Are you one of the top three suggestions? | AI trusts you enough to lead with you |
| Cited | Is your own domain listed as a source? | Your content was used directly, even if a rival got recommended |
The full method behind computing these three yourself, including how many questions to ask and how to exclude anchored ones, is covered in how to measure AI visibility: named, recommended, cited. This guide focuses on turning that into an assessment you can repeat and trust.
By fixing the question set and the engines before you run it the first time, then reusing both every time you re-check. A benchmark that changes its questions between runs can't tell you whether your number moved or the questions did. The Reflexa AI Visibility Index asks 100 fixed buyer questions per category, three times each across ChatGPT, Claude and Perplexity, for 900 observed answers per edition, and reuses the same question set edition over edition for exactly this reason.
Bottom line: the benchmark is the fixed question set, not a number someone else computed on different questions.
The actual AI answer and its sources, for every question you counted, because a number you can't read back to its source can't be acted on. If a tool reports "62% visibility" with no list of the questions asked and no transcript of what each engine said, there is nothing to fix from it. Reflexa's free check keeps the raw answer and sources next to every finding so a low count traces back to one specific sentence an engine gave a real buyer.
None of this requires buying anything, and each step takes a few minutes:
Bottom line: you can get a first, honest read on all three counts yourself in under 20 minutes, before paying for anything that calls itself a score.
There is no industry-standard definition. Most tools mean some blend of how often AI names your brand, recommends it, and cites your domain, compressed into one number. Reflexa does not compute one, because blending those three hides which one actually moved.
Ask 10-20 real buyer questions in your category across ChatGPT, Claude and Perplexity, three times each, and record three things per answer: was your brand named, was it in the top three recommendations, and was your domain cited as a source. Divide each count by the number of eligible answers.
Yes, if you reuse the same fixed set of questions every time you re-check. Changing the questions between runs makes any month-over-month movement meaningless, since you cannot tell whether the count changed or the question set did.
A real benchmark compares the same fixed questions across the same engines over time, with the raw answer kept as evidence next to each count. A benchmark that only shows a single score with no question list and no answers behind it cannot be checked or acted on.
Because a number that blends a blocked crawler with a slow page or a missing FAQ tells you nothing about any of them. The check returns 35+ structural findings, worst first, each with the evidence and the fix, and a separate audit of named, recommended and cited answers.
The free check reads your pages the way ChatGPT, Claude and Perplexity read them, 35+ structural checks, evidence attached, no card needed.
The product: Reflexa Agent · how it works · på svenska