Someone with a villa in Emirates Hills asks who should renovate it. Below, every answer four assistants gave on 15 September 2026, in both languages. Switch between them. Whichever one your customer happened to open decided who they called.
“Who are the best villa renovation companies in Dubai?”
ChatGPT, gpt-5.6-sol, 10:34 UTC, 15 September 2026. Names as it wrote them, in its order. Built from thebigfitout.com, rayfitout.com, bestinhood.com, smartrenovation.ae, buildsmith.ae, propertyfinder.ae, algedra.ae, palmerainteriors.com, dda.gov.ae.
You have viewed 1 answer. 8 companies named. Change a switch to compare.
Every host cited by the latest English answer from each assistant, 15 September 2026. A filled square means that assistant cited that host. 34 hosts in all; 4 fed more than one assistant.
Your customers have stopped opening ten tabs. They ask ChatGPT, Gemini, Perplexity or Claude a question, read one answer, and call whoever was named in it. There is no second page of results to be on.
Those assistants build each answer from a handful of websites they happen to trust for that question. If none of them describes your company clearly, you are not in the answer — however good your Google ranking is, and however much you spend on ads.
The industry has given this two names. GEO, generative engine optimisation. AEO, answer engine optimisation. They mean the same work: making sure the sources an assistant reads say the right things about you, so it names you when it matters.
Almost nobody selling this in the Gulf measures it. They publish a page about it and then do ordinary search work. We measure first, in both languages, and only then tell you what to change.
The one most people mean by AI. Searches the web when it decides the question needs it, and now sells a sponsored card under its answers.
Today: Asked the same English question at 10:07 and 10:34, it kept only two names in common with itself. Its Arabic answer cited three sites, one of them a Dubai government page.
Google’s assistant, built on Google Search. Reaches the widest audience through Android and Google’s own surfaces, and its sources are whatever Google ranks.
Today: Named eight companies in English and seven in Arabic, mostly citing other renovation companies’ own blogs. Two English answers two minutes apart shared four names.
Search first, answer second. Every claim carries a citation, and it leans on directories and comparison lists, which is where regional sites get in.
Today: Cited 15 sites in English, more than any other, and its English and Arabic lists shared one company.
Anthropic’s assistant. Searches on demand, summarises roundups and reviews, and hedges more than the others, so it tends to name more companies with less conviction.
Today: Named fifteen companies in English, the most of any assistant, and eight different ones in Arabic, with two in common.
A full check asks 40 of your customers’ questions, three times each, on four assistants, in English and Arabic. The same question asked twice can return different names, so we never ask once.
You tell us what you sell, who buys it, and how those buyers actually phrase the question. We write 40 of them in English and Arabic, and you approve every one before anything runs. No question you would not recognise as a real customer’s.
Each question goes to ChatGPT, Gemini, Perplexity and Claude, three separate times, in fresh sessions. Assistants do not repeat themselves reliably, so asking once tells you nothing. We keep each assistant’s results apart, because they disagree.
We count who was named, in what position, and how they were described. We trace which websites each answer was built from, which is the part that matters: those sources are where the work happens.
We look at your site the way an assistant does — whether it can reach your pages at all, whether your services, prices and locations are stated in a form it can quote, and whether what the rest of the web says about you agrees with what you say about yourself.
Every line has an owner and a done-by date. Some are ten-minute technical fixes. Some are rewrites. Some are getting onto a specific directory or publication that the assistants keep citing in your category. We put them in order of what will move the answer fastest, not what is easiest for us.
If you already sell search or marketing in the Gulf, your clients have started asking about this and you have no way to measure it. Register as a partner and the whole check runs behind your logo — you enter the client, the category and the city, and a finished report comes back in your colours with Surfup nowhere on it.
Per full check, white-labelled. Lower from ten a quarter.
Published range for a standalone AI visibility audit in 2026.
Self-serve. No call with us required after the first one.
Start with one client you are about to pitch. Send the category, the city and their name, and we will run a single-question check free and send you the result the same day — real names, real sources, timestamped.
One real question, put to a live assistant while you watch. The same list fills with your competitors instead.
The names in an answer are chosen by the assistant. ChatGPT now sells the box under the answer to whoever pays, so there are two ways to appear, and only one keeps working.
This is what a competitor’s paid placement looks like.
We don’t sell or place ads. We work on the other side.
Nobody controls what an assistant says, so nobody can promise you a place in the answer, and we never do. What we promise is that you will know exactly where you stand and exactly what to change.