How to measure AI visibility without fooling yourself
Asking ChatGPT one question and screenshotting the answer is not a measurement. Here is how we sample assistants, why we never blend them into one score, and what we do when an answer cannot be read.
AI visibilityThe SEO Rituals team11 September 20262 min read
More of your buyers now ask an AI assistant before they search. So it is reasonable to want to know whether those assistants mention you. It is also very easy to measure badly.
One answer is an anecdote
Assistants are non-deterministic: ask the same question twice and you can get two different answers. A single screenshot tells you what happened once.
So we ask a fixed panel of questions — the ones your buyers actually ask — to each assistant every week, several times each, and we only report a rate when enough answers came back to mean something. Below that, the honest answer is "not enough data yet".
Never one blended score
ChatGPT, Claude and Perplexity answer differently, cite different sources and change on different schedules. A single "AI visibility score" would average three things that are not the same thing, and a rise in one could hide a fall in another.
So every number we show names its assistant. If an assistant is not connected, it shows as not connected — never as zero, because zero would mean it looked and did not find you.
A parse failure is not a "no"
To tell whether an answer mentions you, a small model reads it. Occasionally that reading fails. The tempting shortcut is to count those as "not mentioned". We never do: an answer we could not read is excluded from the count entirely, and we show both numbers so the difference stays visible. Otherwise a problem on our side would show up as a decline on yours.
Where the useful signal actually is
The most actionable thing is not the mention rate — it is the sources the assistants cite. If a review site or a comparison article shows up in a fifth of the answers about your category, and you are not on it, that is a specific, finite job. We turn each of those gaps into a task with a written playbook.
What we will never claim
We never promise that an assistant will start citing you. They change their behaviour without notice, and a result observed once may not repeat. When a recheck shows you newly mentioned after some work, we report exactly that: observed after the work.