How to check if ChatGPT recommends your business

What you are actually testing

"Does ChatGPT know my business?" hides two questions. Can it describe you when someone types your name? And does it name you to a stranger who asks for, say, an AC repair company in Al Barsha? Only the second brings in new customers, and it is the one this check is built around.

This check covers both: unbranded questions first, then the branded check. You need a browser and a spreadsheet, nothing paid. The quick pass takes about ten minutes; a pass you can act on takes about an hour over two days.

Step 1: write five to ten buyer questions

Write them the way customers type or say them, not the way you would write a keyword. Your WhatsApp enquiries, front-desk phone calls and Google Search Console queries are the best source. Most small businesses have between ten and thirty genuinely distinct buying questions, and a single garage might have as few as six (our small-business research), so five to ten covers the core.

Mix the types of question:

• Best or recommended: "Which garage in Al Quoz is good for a pre-purchase inspection?"
• Urgent need: "Who can fix my AC today in Mirdif?"
• Price: "How much is a deep clean for a two-bedroom flat in JLT?"
• Comparison: "Is it cheaper to hire movers or rent a truck from Dubai to Abu Dhabi?"
• Fit: "Which ladies' salon in Sharjah does keratin and stays open late?"

Two rules: no business names in the questions, yours or a competitor's, and always include the area, as customers do. OpenAI itself suggests adding your city or neighbourhood when ChatGPT's local results come back for the wrong place (OpenAI).

If a real share of your customers ask in Arabic, write two or three questions in Arabic too, the way they would type them: أفضل شركة صيانة مكيفات في دبي (best AC maintenance company in Dubai) or ورشة سيارات موثوقة في القوز (a trustworthy car workshop in Al Quoz). Google lists Arabic among the supported languages for both AI Overviews and AI Mode (AI Overviews, AI Mode), so score your Arabic questions as separate rows.

Step 2: run them in six places, signed out

Customers don't all use ChatGPT. Open all six in tabs in a private browser window and paste each question into every tab. What each does with your question, from its own help or product pages:

Three habits make the result worth trusting:

1. Stay signed out, or start clean. Signed-in history can personalise answers, especially if you have been asking about your own business. Ask signed out where the assistant allows it; otherwise use its incognito or new-chat option.
2. Ask from where your customers are. Run it from the UAE with any VPN off. OpenAI notes a VPN can change the location ChatGPT infers.
3. Expect the answer to move. Ask each question at least twice more, ideally on a different day. Google's own advice for AI Mode is to ask several versions of a question, and in our daily tracking the same prompt can return different sources on consecutive days (UAE measurement report). You are looking for a pattern, not a verdict from one answer.

Step 3: score the answers

For every answer, record whether you were named, who was named instead, and which sources were cited. The last is the column people skip and the one that tells you what to fix: the pages an engine cites are the pages it trusts for your category.

Score each cell 2 if you were named in the answer with the right details, 1 if you were named with a wrong detail (area, phone, price, services) or appeared only in the source links, and 0 if you were not named. Leave a cell blank, not zero, when Google shows no AI Overview for that search.

Add up each column and each row. A column of zeros shows which engine to work on. A row of zeros shows a question you are losing everywhere, and its last two cells show who is winning it and why.

Step 4: the branded check, and why it flatters you

Now ask about yourself by name: "What do you know about [business name] in [area]?" and "What are [business name]'s opening hours and phone number?" Check every fact: name, area, phone, hours, services, prices.

Do this last, and don't let it reassure you. A branded question tests whether an engine can look your name up, not whether it recommends you to a stranger. In our UAE AI visibility report, a software brand we track, separate from the report's two UAE sites, was cited at position one on 4 of 6 engines when the prompt contained its name, while the unbranded version of the same question returned zero citations.

The branded check is still useful for one thing: wrong details. If an assistant tells people you are in Deira when you moved to Al Quoz two years ago, it is costing you the customers who do find you.

What your results mean

Not named anywhere

This is a common result, even for businesses that rank well on Google. A UAE car-servicing directory we measure scored 62 out of 100 on traditional search and 3 out of 100 on AI answer visibility the same day, and was cited on none of its unbranded prompts. Assistants name businesses they can corroborate across sources they trust, so look at your "Sources cited" column: those lists, directories and threads are where you need to be present and accurate. Our guide on how to get recommended by ChatGPT takes the fixes in priority order.

Named by some engines, not others

Normal, and useful: engines differ a lot in how widely they read. In our three-week citation study, of 25 outside sites cited at least once, Perplexity drew on 23 and Claude on 22, Google AI Overviews on 17 and Grok on 15, while Gemini used 8 and ChatGPT only 3 (where AI answers come from). That study covered one software category, so read it as a direction rather than a rule for your trade: third-party work tends to show up in Perplexity and Claude first, and ChatGPT tends to move last.

If you are missing only from Google's AI answers, start with your Google Business Profile and site basics. Google says the same SEO practices apply to AI Overviews and AI Mode, and specifically mentions keeping Business Profile information up to date (Google Search Central). Our Google Business Profile checklist goes field by field.

Named, but with wrong details

Fix this first: a wrong recommendation sends a ready buyer to the wrong area or a dead number. The wrong detail usually lives somewhere the engine reads, such as an old directory listing, a stale Business Profile or an out-of-date page on your own site. Correct it everywhere, in the same wording, and re-run the branded check a few weeks later.

When to stop doing this by hand

By hand, a proper pass takes about an hour each time, and a few runs per engine is a small, noisy sample. Once you are making changes and want to see week by week whether they worked, automate it.

The tool we recommend for this is Cituna. It runs your buyer prompts daily across seven engines (ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode) and connects to Google Search Console, so your AI answers and your Google search data sit in one place.

Starter costs $39 a month for one brand and 10 buyer prompts, with daily scans and fix recommendations. Pro, at $119 a month, covers three brands and 30 prompts and adds alerts and competitor tracking. Where it is weaker: 10 prompts at entry is tight, and a clinic or garage with several services can outgrow it quickly. The free trial needs a card, lasts three days and covers one brand and five prompts: enough to see the setup, not enough to see week-to-week movement. If you want to compare options first, our roundup of AI visibility tools covers the alternatives.

Questions people ask

Sources

• OpenAI Help Center, Searching the web with ChatGPT: automatic and manual search, signed-out web search, citations, approximate location from IP address, memory and VPN effects (read 24 September 2026).
• Gemini Apps Help, Use Gemini Apps: text questions without signing in (read 24 September 2026).
• Gemini Apps Help, View related sources from Gemini Apps: Sources button, not every answer has sources (read 24 September 2026).
• Perplexity Help Center, How does Perplexity work?: real-time web search, numbered citations (read 24 September 2026).
• Claude Help Center, Enable and use web search: web search toggle, automatic search in the newer interface, citations (read 24 September 2026).
• Claude Help Center, Use incognito chats: incognito chats do not use or add to memory (read 24 September 2026).
• Grok product page, x.ai/grok: web, iOS and Android; live web and X search with citations; memory across chats (read 24 September 2026).
• Google Search Help, AI Overviews in Google Search: when they appear, UAE and Arabic availability (read 24 September 2026).
• Google Search Help, AI Mode in Google Search: how to reach it, query fan-out, asking several versions, UAE and Arabic availability (read 24 September 2026).
• Google Search Central, AI features and your website: no special requirements, different models for AI Overviews and AI Mode, Business Profile information (read 24 September 2026).
• Cituna, pricing: engines, plans, prompts, trial terms, Search Console integration (read 24 September 2026).

Drafted with AI assistance from our own research and the sources linked above, and reviewed by the High Street team before publishing. Prices, features and policies change; check the linked source before you act.