Why AI Recommends Your Competitor and How to Fix It in Dubai

AI assistants recommend the business they can read, verify and match to the question. When ChatGPT or Perplexity names your competitor, their information is usually easier to find, clearer to parse and confirmed by more outside sources than yours, regardless of who runs the better business.

You can check each of these reasons yourself, and most of them can be fixed. Below are six common causes, a table to diagnose your own case, and the questions to put to whoever handles your SEO.

How an assistant builds a recommendation

Someone might ask "best physiotherapy clinic in JLT" or "who rents a Lamborghini Urus in Dubai Marina". The assistant answers from two sources.

Training data. This is what the model absorbed before its cutoff: websites, directories, reviews and articles. It changes slowly.
Live search. ChatGPT search, Perplexity and Google AI Overviews fetch pages at the moment of the question, read them and summarise.

In both cases the model looks for three things:

a business that matches the category and the area;
facts it can quote, such as address, hours, prices and services;
signals that other people trust it.

If your site gives the model none of these in readable form, it moves on to the next result.

For the wider picture of how this selection works, see GEO: how AI assistants choose who to recommend.

Reason 1: The assistant cannot read your site

This is the cheapest cause to fix. Typical blockers:

Bots blocked. Some hosting setups and security plugins block crawlers by default. Savage Media checked 537 Dubai restaurant and venue websites on 13.10.2026. In that study, 52 sites blocked bots outright.
Key information inside a PDF or image. The same study found that 147 of 409 fully analysed sites (35.9%) published their menu only as a PDF. A model can sometimes read a PDF, but a price list on a normal web page is far easier to quote.
Content that appears only after JavaScript runs. If the page source is empty and everything loads later, simpler crawlers see a blank page.
The site is down. 55 of 537 sites in that study (10.2%) did not open at all.

Check your robots.txt for lines that block these crawlers:

OAI-SearchBot (ChatGPT search)
PerplexityBot
GPTBot (OpenAI training)

Blocking the training bot is a business decision you can make. Blocking the search bots removes you from live answers. Full figures are in the Savage Media study.

Reason 2: Your competitor states facts plainly

Models quote sentences. A homepage that says "Experience luxury redefined" gives the assistant nothing to quote. Compare a page that says "Physiotherapy clinic in JLT Cluster D, open 8:00 to 21:00, sports injury sessions from AED 350". That one sentence answers three questions.

Look at your homepage and service pages. Could someone copy one sentence and know what you do, where, and roughly for how much? If not, the assistant has the same problem.

Facts worth stating in plain text on the site:

Category and specialisation, in the words customers use
Area and full address, plus which areas you serve
Opening hours, including Friday and Ramadan changes
Price ranges or starting prices
Languages you serve customers in
Who runs the business, with a name

Reason 3: Other sites talk about them, and nobody talks about you

An assistant is more confident recommending a business that several independent sources describe in the same way. Your own site counts as one source. The model also weighs:

Google Business Profile: category, reviews, photos, answered questions
Reviews on Google, TripAdvisor, Talabat, Zomato and similar platforms in your niche
Local directories and industry listings
Press mentions, interviews, guest articles, podcast pages

Picture a competitor with 400 reviews who appears in three "best of Dubai" lists, while you have 40 reviews and no mentions. The model sees less risk in naming them.

Start with the profile you control: Local SEO in Dubai: Google Business Profile checklist.

Reason 4: They have a page for the question, you have one page for everything

People ask assistants specific questions, for example:

"shisha catering for a private event in Abu Dhabi"
"how to cancel a car rental booking in Dubai"
"lawyer for a labour dispute in the UAE in Russian"

The business with a page that answers that exact question has a far better chance of being quoted.

When I built the Moon Mars site, the structure followed the questions buyers ask:

separate service pages for shisha outsourcing, catering, consulting and technology;
city pages for Abu Dhabi and Doha;
33 articles in each language;
a catalogue of 133 shisha lounges.

The Lucent Legal site goes further, with 213 guides in Russian and 213 in English, each answering a single legal question. You can see both in the Moon Mars case and the Lucent Legal case.

Your version does not need hundreds of pages. Write down the ten questions customers ask on WhatsApp before they buy. Any question your site does not answer with a page is one your competitor can win. If answering them means changing the site structure, that is part of website work.

Reason 5: Your details do not match across the web

Say your Google profile says "Al Barsha", your website says "Barsha Heights" and your Instagram bio lists an old phone number. The model sees three slightly different businesses. It may merge them wrongly or trust none of them.

Common mismatches in the UAE:

The trade name on the licence differs from the brand name customers know.
English, Arabic and Russian versions spell the name differently.
Old branch addresses still live on directories.
Phone numbers switch between a landline and a WhatsApp mobile.

Pick one version of the name, address, phone and hours, and make every profile match it. On multilingual sites, link the language versions with hreflang so the model knows the EN and RU pages describe the same business.

Reason 6: No structured data, so the model has to guess

Schema markup tells machines, in a fixed format, what your page is about: business type, address, hours, price range, reviews, menu. Assistants and search engines can read a page without it, but with it they misread less.

The Savage Media study found Restaurant schema on 90 of 409 venue sites (22.0%). It also found that 22.7% had no meta description at all. A competitor who has both gives the model cleaner input than most of the market.

At minimum, ask for this markup:

LocalBusiness, or the specific subtype: Restaurant, MedicalClinic, AutoRental or LegalService
address and geo coordinates
opening hours
a link to your Google profile

Diagnose your case in 20 minutes

Symptom	Likely cause	How to check	Fix
AI says it has no information about you	Site blocked or unreadable	Open `yoursite.com/robots.txt`; view page source	Allow search bots; render content server-side
AI describes you with wrong details	Mismatched data across profiles	Compare Google profile, site footer, Instagram bio	One version of name, address, phone, hours everywhere
AI names you for broad queries, competitor for specific ones	Missing question pages	List ten customer questions; search your site for each	One page per real question
AI calls the competitor "popular" or "well reviewed"	Fewer reviews and mentions	Count reviews and third-party listings for both	Ask every satisfied customer for a review; get listed
AI quotes your competitor's prices but not yours	Prices hidden in PDF or not published	Search your site for "AED"	Put starting prices on web pages
AI mixes you up with another business	Weak entity signals	Search your brand name in quotes	Schema, consistent naming, an About page with a real person

To run the first step systematically, follow how to check what ChatGPT says about your business.

Hypothetical example: two clinics in Dubai Marina

Hypothetical example: two aesthetic clinics sit in the same tower.

Clinic A has a polished site built as one long page and prices in a downloadable PDF. Its Google profile has 35 reviews and the category "Medical spa".
Clinic B has a plainer site with a page per treatment and starting prices in AED on each page. A named doctor appears on the About page. Its Google profile has 280 reviews and the category "Aesthetic clinic".

A user asks Perplexity for "aesthetic clinic in Dubai Marina with transparent prices". The assistant finds Clinic B's treatment pages, quotes the price, cites the reviews and names the doctor. Clinic A appears, if at all, as "another option nearby". Its doctors may be as good, but the assistant had less to read about Clinic A and nothing to confirm it with.

What not to do

Do not stuff pages with "best clinic in Dubai". Models summarise meaning. Repeated superlatives add nothing quotable and can read as spam.
Do not buy fake reviews. Platforms remove them, and a sudden spike looks suspicious to people and machines alike.
Do not block all bots to "protect content" until you know which ones feed live search.
Do not publish 50 thin AI-written articles. Pages that repeat each other with different city names give the model no new facts.
Do not hire anyone who guarantees ChatGPT will recommend you. Nobody controls the model's answer. A contractor who promises it will likely report the rest of the work the same way.
Do not rewrite your brand name in every language. Transliterate it once and keep that spelling everywhere.

Questions to ask your SEO contractor

Put these to your current agency or any new one:

Which AI crawlers does our robots.txt allow, and why?
Which ten customer questions does our site answer with a dedicated page, and which does it miss?
What schema types are on our homepage and service pages? Show me the validator result.
Where do our name, address and phone differ across the web today?
How will you measure change in AI answers: which prompts, which assistants, how often?
Do we have an llms.txt, and what is in it? (Background: llms.txt: what it is and how to write one.)

Good answers name specific files, pages and prompts. "We'll boost your visibility" does not answer any of the six.

Checklist

Open robots.txt and confirm OAI-SearchBot and PerplexityBot are allowed.
View your homepage source and confirm the main text is there without JavaScript.
Move menus, price lists and service lists from PDF onto web pages.
Write one plain sentence on the homepage: what you do, where, from what price.
Make name, address, phone and hours identical on the site, Google profile, Instagram and directories.
Set the most specific category in Google Business Profile.
Create a page for each of the ten questions customers ask before buying.
Add LocalBusiness schema, or the right subtype, with address, hours and geo coordinates.
Add an About page with a real person's name and role.
Ask every satisfied customer this month for a Google review.
Ask ChatGPT, Perplexity and Google the same five questions about your category and area, save the answers, and repeat in 30 days.

Questions and answers

Can I pay to be recommended by ChatGPT?

Nobody can sell you a place in ChatGPT's organic answers. You can only improve what the model reads about you: your site, your Google Business Profile, reviews and mentions on other sites. Anyone who promises a guaranteed recommendation is selling something nobody controls.

How long does it take before AI answers change?

Assistants that search the web live, such as ChatGPT search and Perplexity, can pick up changes once your pages are crawled and indexed. Answers that rely on training data change only when the model is retrained, so they lag behind. Check monthly with the same set of questions to see movement.

Does llms.txt make AI recommend me?

No single file does that. llms.txt gives AI tools a short, clean map of your site and key facts, which helps them read you correctly. It works on top of crawlable pages, clear facts and outside mentions, and does not replace them.

My competitor has a worse product. Why does AI still pick them?

The model cannot taste your food or test your service. It weighs what is written about you and how easy it is to confirm. A weaker business with clear pages, consistent details and more reviews often looks safer to recommend.