It usually starts with a screenshot. A partner, a client or a curious owner asks ChatGPT for the best firm in their category in Dubai, and the answer names three competitors, describes them confidently, and does not mention you at all. The website is fine. The Google rankings are fine. The assistant has simply never heard of you, or has heard and decided not to say.
Quick answer: when ChatGPT leaves a business out, the cause is almost always one of ten things, and they fall into three layers. Can it read you? (A blocked crawler, a site missing from the search indexes it uses, or answers locked inside images.) Does it find the answer? (No page answers the question it was asked, or the question was asked in Arabic.) Does it trust you? (A name that means two things, details that disagree across the web, nobody else mentioning you, no presence on the map, or old facts still in circulation.) Test properly first, then work through the checks in that order; the first three take minutes to days.
This is the diagnosis we run before quoting anyone for AI search work, written so that you can run it yourself. Each check comes with a test you can do today and the fix, and a handful will tell you within an hour whether the problem is technical or reputational.
First, test it properly
A single screenshot proves very little. ChatGPT decides for itself whether to search the web for a question, the wording changes the answer, earlier conversation and memory colour it, and for local questions OpenAI says it may use an approximate location based on the user's IP address. So test the way a stranger would meet you: in a temporary chat, logged out of anything personalised, with each question phrased three ways and asked more than once.
| Ask | What it tests | If you are missing or wrong, look at |
|---|---|---|
| "What is [your business] in Dubai?" | Whether it knows you exist, and as whom | Checks 1, 2, 5, 10 |
| ★ "Best [your category] in [your area]" | Whether it considers you at all | Checks 6, 7, 8 |
| "How much does [your service] cost in Dubai?" | Whether your pages answer real questions | Checks 3, 4 |
| "[You] or [your main competitor]?" | What it believes about you | Checks 6, 7, 10 |
| The second question again, in Arabic | Whether you exist in the other language | Check 9 |
Write down what each answer says and which sources it shows. The sources matter as much as the names: they are the pages the assistant trusted instead of yours.
Check 1: You have blocked the reader
OpenAI's crawler documentation is blunt: sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers". Our comparison of ChatGPT and Google Ads explains what each of OpenAI's crawlers does; the check here is simply whether anything on your stack stops the search one.
| Where it gets blocked | What to look for |
|---|---|
| robots.txt | A Disallow for OAI-SearchBot, or a blanket rule written for "all AI bots" during a nervous week |
| ★ Cloudflare | Security Settings, "Configure AI bot policies": Search should be allowed. Since 15 September 2026 new domains block Training and Agent bots on pages showing ads, and Search stays allowed, but older zones keep whatever someone chose earlier |
| Firewall and country rules | Bot protection or "under attack" modes that challenge every automated visitor |
| Security plugins and hosts | WordPress security plugins and managed hosts with their own bot lists |
| Server logs | Requests from OAI-SearchBot answered with 403 or 429, or no requests at all |
Two details from OpenAI's own page are worth knowing. A robots.txt change takes about 24 hours to reach its systems, so do not retest the same afternoon. And its separate ChatGPT-User agent, which fetches a page when someone in a chat asks about it, is "not used to determine whether content may appear in Search", so allowing it is about letting people who already found you read you, not about being found.
Check 2: You are missing from the indexes it searches
When ChatGPT searches, OpenAI says it "leverages third-party search providers, as well as content provided directly by our partners". It does not publish the full list, which makes the safe course simple: be properly indexed in both Google and Bing. Microsoft's Copilot answers from Bing's index in any case, and Google is explicit that a page needs to be indexed and eligible for a snippet to appear in its own AI Overviews and AI Mode.
The test takes ten minutes. Search site:yourdomain.ae on Bing as well as Google, and look for your money pages, not just the home page. Then check for a stray noindex, a nosnippet tag, or a canonical pointing somewhere else on the pages that matter. Plenty of Dubai sites have Search Console set up and have never opened Bing Webmaster Tools, which is free and takes the same sitemap.
Check 3: The answer is not on the page in words
Assistants quote text. A price list in a JPEG, a service menu in a PDF, a spec sheet inside a carousel, or content that appears only after a script runs is, for practical purposes, not on the page. Google's own guidance for its AI features makes the same point in a sentence: make sure that "important content is available in textual form".
The test: open your key pages, turn on the browser's reader view or view the page source, and read what is left. If the prices, the service areas, the opening hours or the qualifications have vanished, so have they for the assistant. The fix is unglamorous copy work, and it is usually the single biggest improvement on a design-led site.
Check 4: No page answers the question being asked
People do not ask assistants for "interior design services". They ask how much a villa fit-out costs in Dubai Hills, whether mainland or free zone is cheaper for a consultancy, which clinic does a particular treatment on a Friday. If no page on your site answers a question like that in its first few sentences, the assistant answers it from a page that does, and names whoever wrote it.
List the ten questions your sales team hears most and put a URL next to each. Every blank is a page to write, and the questions about cost and comparison are the ones most worth writing first because almost nobody in a category publishes real numbers. How to build those pages is a job in itself; the diagnosis is simply the count of blanks.
Check 5: Your name means two things
Assistants build a picture of an entity from everything written about a name. If your trade name is a common word, is shared with a business in another emirate or country, or differs from the name on your licence, the picture blurs. The symptom is distinctive: ask "what is [your business]?" and the answer describes somebody else, or merges two of you.
The fix is to give the name context everywhere it appears: the same full name with the city and the category, Organization or LocalBusiness structured data on the site with sameAs links to your official profiles, and an About page that states plainly who you are, where you are licensed and what you do.
Check 6: Your details disagree across the web
Models trust facts that several sources agree on. Five different phone numbers, two addresses and three versions of the business category tell an assistant that nobody is sure.
| Fact | Where it must match |
|---|---|
| Name, exactly | Website, Google Business Profile, Apple Business, LinkedIn, directories, invoices |
| Address and area | Website contact page, map listings, directories, structured data |
| Phone and WhatsApp | Everywhere a customer can see it |
| What you are | Map category, website headline, directory category, LinkedIn industry |
| ★ Prices and services | Website, listings and anything a partner publishes about you |
Google adds one condition worth copying for every engine: structured data should match the visible text on the page. Markup that claims a price or a rating the page does not show is worse than none.
Check 7: Nobody else vouches for you
An assistant asked for the best firm in a category leans on pages that compare firms: industry lists, press coverage, association directories, review platforms, "top ten" articles that someone in the trade bothered to write. A business that exists only on its own website has one source saying it is good, and that source is the business. The test: search your category plus "Dubai" and read the first two pages of lists and articles. Count how many mention you. Then look at the sources ChatGPT cited in your test answers; those are the publications to be in.
This is the slowest check to fix and the one that moves the most. It is earned through digital PR, published data people want to quote, memberships, partnerships and reviews, not bought, and it is why our case below took months rather than weeks.
Check 8: You are local but invisible on the map
For "near me" and "best in [area]" questions, assistants lean on map and listing data, and BrightLocal's survey shows that people now ask them exactly those questions. A business with no verified Google Business Profile, no Apple Business place card, a handful of reviews or a pin in the wrong building is starting a local question from behind. Our guide to getting on Google Maps in Dubai covers the set-up; the check here is whether the profile exists, matches check 6, and carries enough recent reviews to be worth summarising.
Check 9: Your customers ask in Arabic
Run your best question in Arabic, for example أفضل مكتب محاسبة في دبي ("best accounting firm in Dubai"), and compare the answer with the English one. If you appear in English and vanish in Arabic, the assistant found no Arabic pages of yours to draw on and named whoever had them. A translated page rarely fixes it, because Arabic searchers phrase questions differently; our guide to Arabic keyword research explains why, and the fix is Arabic pages written around the questions as Arabic speakers actually ask them.
Check 10: The web remembers an older version of you
Assistants repeat what they read, including what is out of date: last year's prices on a forgotten landing page, an old address on a directory, a closed branch, a service you no longer offer. Worse, they can attach the right name to the wrong facts. Ask about your prices, locations and services and note anything stale, then fix it at the source: retire or redirect old pages, put a date on pages that carry prices, and state in the first paragraph which emirate, branch or service a page applies to.
Reading your results
Most businesses fail two or three checks, not one, and the pattern of the test answers usually says which. Not named anywhere, for any question, points to the first three checks. Named for your brand but never for your category points to checks 4, 7 and 8. Named with the wrong facts points to 5, 6 and 10. Named in English but not Arabic is check 9. And a competitor named in your place, with their article cited as the source, tells you exactly which page you need to write.
From zero to 52 citations a month: Founders Gate. A Downtown Dubai business setup firm whose partners started asking ChatGPT and Perplexity the questions their clients ask and kept hearing competitors' names. Our first panel run in February 2025 found the firm cited in none of 20 tracked setup questions, with competitors named in 11 answers; the site had no cost figures anywhere and no structured data beyond a bare Organization tag. The work was an honest mainland-versus-free-zone cost index in line-item tables, a native Arabic cluster, schema on the core pages, and digital PR that put the index in Gulf News and Khaleej Times. Perplexity began citing it in late October 2025, and by mid-2026 the firm was earning more than 52 citations a month across the tracked panel. There was a setback worth knowing about: in December 2025 a ChatGPT answer cited the firm for a Sharjah free-zone question with its Dubai prices attached, which took six weeks of entity clean-up to put right. The full account is in the Founders Gate case study.
Worked example: a Business Bay accounting firm. Illustrative. The partners test "best accounting firm for VAT registration in Dubai" and three competitors come back. Check 1 finds a security plugin returning 403 to OpenAI's search crawler, fixed in ten minutes. Check 2 finds the site indexed in Google and barely in Bing; the sitemap goes into Bing Webmaster Tools. Check 3 finds the fee schedule is a PDF; it becomes a page. Check 4 finds no page on VAT registration costs at all; one is written with real figures. Checks 6 and 8 find two phone numbers and a map pin on the wrong tower. Check 7, nobody listing them, is the long one, and goes on the quarter's plan. They retest in a month.
When to get help
Checks 1, 2, 3, 6 and 8 are within reach of any business with a website login and an afternoon. Checks 4, 5, 7, 9 and 10 are where most of the effect is and most of the effort: writing pages that answer real questions in two languages, cleaning up an entity, and earning mentions that other sources repeat. That is the work of our AEO programme, which begins with exactly this test run across the questions that matter in your category and five answer engines, so you see where you stand before you pay for anything else.
Frequently asked questions
Why doesn't ChatGPT recommend my business?
Usually one of three things. It cannot read your site (a robots.txt rule, a firewall or a CDN setting blocks OpenAI's search crawler), it cannot find a page that answers the question it was asked (prices, comparisons and services that exist only in images or nowhere at all), or it does not trust what it found enough to name you (inconsistent details, few reviews, nobody else mentioning you). Test with the same questions your customers ask, in a temporary chat, and work through the checks in that order.
Does ChatGPT use Google or Bing to find businesses?
OpenAI says ChatGPT search uses third-party search providers together with content supplied directly by its partners, and it does not publish a complete list of those providers. The practical answer for a business is to make sure it is indexed in both Google and Bing, which takes an afternoon with Search Console and Bing Webmaster Tools, and to allow OpenAI's own search crawler, OAI-SearchBot, which OpenAI says is what surfaces websites in ChatGPT search.
How long does it take for ChatGPT to start mentioning my business?
Fixing access is fast: OpenAI says a robots.txt change takes about 24 hours to reach its systems. Being named is slower, because it depends on content, consistency and third-party mentions accumulating. In the Founders Gate engagement on our site, the first citations arrived about nine months after the work began and reached 52 a month on a tracked panel by mid-2026.
Why does ChatGPT give wrong information about my business?
Because the web gives it wrong information: an old price on a page you forgot, a directory listing with a previous address, a branch that closed, or a page that does not say which emirate or service it applies to. The fix is at the source, not in ChatGPT: correct or retire the outdated pages, state facts plainly with dates, make every listing agree, and test again after a few weeks.
Does ChatGPT give everyone the same answer about local businesses?
No. Answers vary with the exact wording of the question, with the conversation so far and memory, and, for local questions, with location: OpenAI says ChatGPT may use an approximate location based on the user's IP address to give relevant local results. Test in a temporary chat, phrase each question three ways, run it more than once, and record what you see rather than relying on one screenshot.
Do Arabic pages help a business get mentioned in ChatGPT?
For questions asked in Arabic, yes. An assistant answering in Arabic draws on what it can find written in Arabic, and if the only Arabic content in your category belongs to directories and competitors, those are what it names. Ask your key questions in Arabic as well as English during testing; a business that appears in one language and not the other has found its gap.
Being left out of an answer is rarely a judgement on the business. It is usually a reader that was blocked, a question nobody answered, or a fact nobody else confirmed, and each of those has a fix. If you would rather we ran the ten checks for you, send us your website and the three questions you most want to be named for, and we will come back with what ChatGPT says today, why, and what we would fix first.
