You can measure this yourself, today, for free. The method matters more than the tool — most AI visibility checks test the wrong surface and report businesses as invisible when they are not. Here is one you can defend.
To check your AI visibility in the UAE, ask eight fixed buyer questions on Google AI Mode, Google AI Overviews, ChatGPT with search enabled, Gemini and Perplexity; record the full answer, date and engine for each; and score whether you were named, where you appeared among the businesses listed, and whether the description was accurate. Repeat the identical set monthly.
Not your keywords. The question a person asks when they have a problem and money. The reliable shape is category + area + constraint.
| Weak (keyword-shaped) | Strong (buyer-shaped) |
|---|---|
| dental clinic Dubai | Which dental clinic in Jumeirah is good for a nervous adult patient? |
| Italian restaurant | Where can I take clients for Italian in DIFC that isn't loud? |
| interior fit-out company | Who does office fit-out in Dubai for a 20-person office on a tight timeline? |
| beauty salon Marina | Best salon in Dubai Marina for keratin treatment on damaged hair? |
Write eight to ten. Then freeze the list. The whole value of this exercise is comparability across months; a question set that drifts produces a chart that means nothing.
This is where most checks go wrong. Run each question on:
Use a UAE location and English first; add Arabic if a meaningful share of your customers search in Arabic. Log out or use a private window so your own history does not personalise the result in your favour.
For every question, save four things: the question, the engine, the date, and the complete answer text. A spreadsheet is fine.
date | engine | question | named | position | accurate | answer
2026-07-26 | AI Mode | best Italian in DIFC for clients | yes | 3 of 5 | yes | "..."
2026-07-26 | ChatGPT+search| best Italian in DIFC for clients | no | — | — | "..."
2026-07-26 | Perplexity | best Italian in DIFC for clients | yes | 1 of 4 | no* | "..."
* described us as "casual" — we are fine dining
Keeping the answer text is not bureaucracy. It is the only way to tell, three months later, whether an improvement is real or whether the engine simply phrased things differently that day.
| Signal | Question it answers | What a bad result means |
|---|---|---|
| Named | Did the answer mention us at all? | Legibility problem — the system cannot confidently attach you to the query's constraints |
| Position | Where among the businesses listed? | Competitive problem — you are known but not preferred |
| Accurate | Was what it said about us true? | Consistency problem — your sources disagree, and the engine picked the wrong one |
These have different fixes, which is why collapsing them into one score hides the actionable part. "Named but described wrongly" is usually a fast fix — a stale directory listing or an outdated profile field. "Never named" is structural.
The answer is downstream. If the inputs are broken, no amount of re-asking helps. Check five things:
| Input | How to check it | Common UAE failure |
|---|---|---|
| Google Business Profile | Search your business name in Google Maps | Missing, unverified, wrong category, or a service area so broad the pin lands hundreds of kilometres from your customers |
| Indexation | site:yourdomain.com in Google | Only the homepage indexed; service pages never submitted |
| Structured data | Google's Rich Results Test on your homepage | None at all — in our Dubai pilot, 71% of testable sites had none |
| Entity linkage | Look for sameAs in your page source | Absent, or pointing only at your own homepage, which corroborates nothing |
| Consistency | Compare name, address, phone across site, profile and directories | Three phone formats and two spellings of the same company |
Same questions, same engines, same rough date, one file. Change one thing at a time on the business side so you can attribute movement. Expect noise: a business can be named in one run and not the next with nothing changed, which is precisely why the series matters and the snapshot does not.
Write eight buyer questions in your customer's words, ask them on the surfaces customers use — Google AI Mode and AI Overviews, ChatGPT with search on, Gemini, Perplexity — and record the full answer, date and engine each time. Score whether you were named, in what position, and whether the description was accurate. Repeat monthly.
They draw on different data at different moments. A model answering from memory reflects the web months ago; the same model with search enabled reflects what it retrieves now. Answers also vary between identical runs by design.
No — it is the most common error. A bare API call with no search tool tests what the model remembers, not what a customer sees, and it systematically under-reports local businesses. Always measure the consumer-facing surface.
There is no industry benchmark, so treat any absolute score sceptically, including ours. What is meaningful is your own mention rate across a fixed question set over time, and your standing against the competitors named alongside you.
Monthly. More frequently and run-to-run variance drowns the signal; less frequently and you cannot attribute changes to actions.
Same method, done properly: buyer questions for your category on the consumer surfaces, plus Maps presence, reviews and structured data, scored 0–100 with the raw answers attached. Free, private to you, no obligation.
Get your free AI visibility scan