A measurement protocol, not an opinion. Live logged runs, frozen raw answers, published scoring. Every number in the Index traces back to a stored AI answer.
When a client types a real buying-intent question into an AI assistant — “best hair transplant clinic in Dubai” — which businesses does the assistant name, how often, and in which position? The Index measures that, and only that.
It is not a survey of client satisfaction, not an audit of service or clinical quality, not a review platform, and not a ranking of outcomes. It measures presence across the AI answer surfaces that now sit ahead of how clients discover a business.
Some visibility benchmarks publish modeled estimates from an AI model's own general knowledge, without ever running live queries against the systems they score. This Index does not model.
Every issue is computed from live, logged, timestamped responses collected from the named AI systems during the stated measurement window. Raw responses are frozen in our database at collection time and kept as evidence. Any published figure — any score, any rank, any percentage — traces back to the stored answers that produced it.
The composite score is computed from three assistants, from a UAE environment. Each is captured the way a client actually experiences it — which, deliberately, is not the same method for all three:
Captured from the real consumer app through a live browser session. We do not use the ChatGPT API here: its output differs structurally from what users see in the app, so the API would misrepresent what a client is actually shown.
Via the official API, whose output matches what users see in the app.
Via the official API, whose output matches what users see in the app.
Engines not named here are not part of the current methodology. When a new engine is added, this page is updated and the version incremented.
Each issue runs a fixed battery of neutral, buyer-intent questions for one procedure in one city — the questions clients actually ask, in English.
Terminology is strict: a “query” is a question we ask — always 20. A “sampled response” is one AI answer to a query — 180 in total. Both values are pulled live from the issue manifest — prompt_count and valid_observations — never hardcoded.
Presence rate is the core signal: the share of observed answers in which a business appears at any position. Mentions are position-weighted — a business named first counts for more than one named last — and averaged across all prompts and repeats, then combined across engines using the weights above into a composite AVI from 0 to 100.
For this issue, scoring is computed over 180 sampled AI responses. The composite is a period benchmark tied to the stated measurement window — it answers “how visible was this business to AI assistants during these dates,” nothing more.
AI answers are probabilistic: the same question asked twice can return a different list. That is why the Index never publishes a single screenshot as proof. Every figure aggregates many observed answers, the count is published with every issue (180 this issue), and per-engine differences stay visible in the data instead of being hidden inside one number.
Scores are calculated automatically under this methodology. Commercial relationships do not influence scores or positions. No business — and not the publisher — can hand-edit a rank. Disputes are re-run publicly and corrected in the next issue.
A verified representative of any listed business may request:
If an entry contains an error (name, attribution, matching), write to marsel@trygrandbazar.com with proof of authority to represent the business.
We re-run the standard battery and publish the refreshed figures with a new window date.
On formal written request, we remove the business from public Index pages. Delisted businesses may still be counted in aggregate, non-identifying statistics.
We acknowledge every request within 5 business days. Where re-measurement confirms an error caused by our data collection or processing, we correct the entry and publish a dated correction note on the affected page.
Every issue and every row carries its measurement window. The Index is re-measured from scratch every Thursday. This page is versioned; any change to engines, prompt counts, weights, or scoring is logged below and increments the version.
Questions about your AI Visibility Index? Drop your email and ask — we usually reply fast.