How we measure AI visibility
The whole method, in the open: where the questions come from, how many times they are asked, what gets counted, and how the score is built.
Tabyian is an AI visibility measurement platform. It measures how often a business is named in the answers ChatGPT, Gemini and Perplexity give to the questions its customers actually ask. This page sets out exactly how the measurement is taken, so you can judge the number before you believe it. The same method is applied to businesses across Jordan, Saudi Arabia, the UAE, Qatar, Egypt and the wider Gulf.
The rule everything else serves
No language model invents a number in a Tabyian report. Deterministic code measures, counts and records; models are used only to put what was measured into readable prose. A separate review pass then checks every sentence against the recorded figures, and prose that contradicts a figure is held for a human rather than sent to a customer.
72 questions, written the way a buyer writes them
We write 72 buying questions for your category and your city, in the dialect of the market you sell in — not polished standard Arabic, but the way people actually type on a phone. They are spread across nine buying intents: best, price, comparison, cheapest, nearest, recommendation, problem, logistics and urgency.
One question in nine is written in English, because buyers across Jordan, Saudi Arabia, the UAE, Qatar, Egypt and the wider Gulf mix the two languages. Your business name never appears in any question. The measurement works by looking for your name in the answers, so a name in the question would come back in every answer and every "mention" would be an echo of our own asking.
Three engines, three rounds
Every question is put to ChatGPT, Gemini and Perplexity, three times each. The three rounds are not padding: an assistant does not give the same answer twice, and a measurement taken from a single ask measures luck rather than visibility.
72 questions × 3 engines × 3 rounds. Every answer that comes back is recorded verbatim and dated, and the count is taken over what actually arrived — not over what was planned. If an engine refuses a batch of requests, the report says so rather than hiding it in the denominator.
What gets counted
Mentions of your name. Whole-token text matching, never substrings, across the spellings your name really appears under: Arabic forms, Latin transliterations, with and without the definite article. And if your business name is also an ordinary word — "Swift", or the Arabic «سريع» — the bare word is not counted on its own. Only the unmistakable forms are, because counting the ordinary word credits you with mentions that never happened.
Who was named instead. We read the answers, extract the businesses named in them, and count those by the same deterministic method. You do not have to name your competitors — the answers name them.
The sources the engines cited. We collect every link the answers leaned on and check a sample of them for your name or your domain. An assistant builds its answer out of pages that exist on the web, and knowing which pages beats any amount of guessing.
Reading your site as the crawlers read it
We fetch your site the way the AI crawlers fetch it — not as a browser renders it for a visitor — and record what they find and what they do not: robots.txt and which bots are allowed, structured data, how much readable text the page actually serves, and whether your service names, price and contact details exist as text at all.
The score out of 10
The score is the sum of five published components, each of which states its own reason in words: mention rate (3 points), crawler access (2), readability (2), answer content (2) and third-party presence (1). Alongside it sits an achievable ceiling showing what specific fixes could recover.
What we do not do
We do not promise a commercial outcome and we do not tell you your sales will rise — we do not measure that and cannot. We make no claim about your site before we have fetched it. Every figure in the report comes from a recorded, dated answer, and a verbatim quote sits next to it.
Do you see the questions?
Yes. All 72 appear in the report exactly as they were asked, with the recorded answers. Before buying you can try the free check: twelve questions across two engines, showing the count and the questions themselves.