To find out whether AI recommends your business, ask it to help you choose a company that provides your service—without mentioning your company’s name.
That distinction matters. “What do you know about my company?” gives the AI a business to discuss. “Which companies can help me with this?” asks it to find options. If you want to know whether a potential customer could discover you through an AI recommendation, start with the second question.
You can run a useful first check yourself. Choose a few realistic customer questions, try them in an AI assistant with web search, and look at which businesses it suggests. What matters is whether your company appears as a relevant option for the work you want to win.
Ask as a customer, not as the business owner
Start with one service you want to attract customers for. Then think about what someone would need to tell an assistant to get a useful recommendation: the job, the location and any requirements that could affect their choice.
Suppose you run a video production company in Dubai. Asking “What is the best video company?” leaves a lot open. A restaurant looking for short social videos has a different brief from a company commissioning a large corporate film.
More useful questions might be:
- “Which companies in Dubai can produce short social media videos for a restaurant?”
- “Can you recommend a Dubai video production company for an English and Arabic product launch video?”
- “I need a team to film customer interviews at our Dubai office. Which companies should I consider?”
These are illustrative questions, not a list every business should copy. Use the kinds of requests that come through your own enquiries. If customers regularly ask about a particular area, language or type of project, include that detail where it belongs.
Keep the questions natural. Do not build an unusually specific request around everything that makes your company distinctive just to encourage it to appear. You want to learn what happens when someone looks for a suitable supplier, not whether you can steer an answer towards yourself.
Start with a manual check in an incognito window
For a first look, you do not need a monitoring subscription. A browser and a simple spreadsheet will do.
In Chrome, open the three-dot menu and select New Incognito window. You can also press Ctrl + Shift + N on Windows or Command + Shift + N on a Mac. Incognito starts a separate browsing session; stay logged out of ChatGPT for this check so you do not reconnect it to your usual account. Chrome’s incognito guide.
Then follow these steps:
- Open ChatGPT and check that you are logged out. If the service requires you to sign in, record that you are testing a signed-in experience instead.
- Start a new conversation and select Search from the tools menu, if available. If search is unavailable, do not mix that response with your web-search results.
- Paste one customer question without your company name. For example: “Which companies in Dubai can produce short social media videos for a restaurant? Please include sources.”
- Save the complete answer and its links. Do this even if your company is absent.
- Start another fresh conversation for the next question. Do not follow up with “What about my company?” and count the resulting answer as an unsolicited recommendation.
OpenAI’s guidance confirms that ChatGPT web search is available to logged-out users, subject to usage limits. Interface and access may vary, so record what you actually used.
Incognito is useful for separating this check from your normal browser session. It does not make you anonymous or change your IP address by itself. Your location can still affect the experience, and staying logged out does not reproduce every customer’s situation. Think of it as one clearly recorded way to check, rather than a universal view of AI recommendations.
For a Dubai service, write Dubai—or the relevant neighbourhood—in the question rather than relying on “near me.” If you serve Arabic-speaking customers, keep a separate set of equivalent Arabic questions and have a fluent speaker check that they express the same need.
Look for a recommendation, not just your name
Read how the answer presents each company. There is a meaningful difference between being suggested as a supplier and appearing somewhere in the supporting material.
If your business is listed among companies the customer should consider, with an explanation of why it fits, that is a recommendation in that answer. If your website is cited as the source of an industry fact but your business is never suggested, you have been cited—not recommended as a provider.
An incidental mention is different again. The assistant could refer to a company while discussing an example or comparing services without presenting it as an option for the customer.
When your business does appear, check the fit. A recommendation for a service you do not provide is not the result you were looking for. For the hypothetical video company, being suggested for wedding filming would be of limited use if it only takes on commercial projects.
Also confirm that the answer refers to the right business. Open any company links and read the explanation. A familiar name alone is not enough when more than one business could share it.
Try several customer questions, then repeat them
One answer gives you an observation. A small set of questions gives you more to work with.
For a manageable first check, choose three genuine customer requests and ask each twice in separate conversations. This is a practical starting point, not a statistically representative measure of your market. Decide on the questions before you begin and keep every answer, including those where your business does not appear.
A simple note for each response is enough:
| What to record | Why it helps |
|---|---|
| Exact question, date and AI product | Lets you repeat the same check later |
| Was your business suggested as a supplier? | Separates recommendations from other mentions |
| Which service was it recommended for? | Shows whether the recommendation fits your offer |
| Explanation and linked sources | Gives you something concrete to investigate |
Avoid turning a result such as two appearances in six answers into a claim that you reach one-third of customers using AI. You have checked six responses to questions you selected. You have not measured how frequently customers ask those questions or what they see in their own conversations.
The more useful finding may be a difference between services: your company appears for one kind of work but not another. That tells you where to look more closely.
When to use an API or a tracking application
Manual checks let you see the answers for yourself. They become time-consuming when you want to follow several services, languages and AI products regularly. At that point, you have two other routes.
An API for a custom, repeatable test
An API lets a developer send a saved set of questions automatically and store the responses. It suits a business with technical support that wants control over the questions, schedule and records.
Give the developer a concrete brief: use the same unbranded questions, keep each request independent of previous conversations, record the model and search settings, and save full answers, citations and failures. If you intend to measure web-backed answers, verify that a search actually ran. OpenAI’s Responses API supports a web-search tool, but simply making a model call does not establish that it searched the web.
Keep API results in their own series. They show how the chosen model and configuration responded; do not label them as a reproduction of what customers see in the ChatGPT app. Agree a usage budget and include the cost of building and maintaining the test before choosing this route.
A tracking application for ongoing monitoring
A dedicated application can be more convenient when you want recurring reports without maintaining your own collection system. As one example, Profound describes a system that sends prompts to answer engines and records their responses. That is a description of its documented approach, not a recommendation based on a hands-on comparison.
Before subscribing to any tool, ask it to demonstrate your own customer questions. Check:
- What it measures: the consumer product, an API, or another collection method—and whether web search is involved.
- What you control: your exact prompts, language and location settings, where supported.
- What you can inspect: full answers, sources, dates and failed checks, rather than only a headline score.
- What the score means: a brand mention, a citation or an actual supplier recommendation, and which responses form its denominator.
- What the plan includes: products covered, prompt limits, repetitions, reporting frequency and total cost for your intended setup.
Register your brand name, domain and relevant spelling variants so you can check whether matches refer to the right business. Keep those identifiers separate from the customer prompts: the tool needs to know what to look for, but the discovery question should not supply the answer.
Choose the simplest route that serves your purpose. Manual checks suit an initial investigation; a tracker can make recurring review easier; an API suits a custom process with technical ownership. None removes the need to read a sample of the underlying answers.
If your company does not appear, investigate before changing anything
An absent name can be frustrating, especially when the answer includes other providers. But it does not explain why those companies were selected, and it does not prove that your business is excluded from future answers.
Start by reading the recommendations. What requirements did the answer focus on? Which sources did it link to? Are the suggested businesses relevant to the question, or did the assistant misunderstand the request?
Then look at your own website from the same customer’s perspective. If the question was about bilingual product videos, can a visitor easily establish whether you offer that service? Is there a relevant example, a clear description of the work and a way to enquire?
This is a useful page review even if you cannot establish why the AI omitted you. It gives you a specific improvement to consider instead of guessing at the selection process. Clearer information may help customers understand the offer; it does not guarantee an AI recommendation. OpenAI states that placement in ChatGPT search is not guaranteed.
If the question does not reflect a service or customer you actually want, revise your test rather than reshaping the business around an irrelevant answer. Keep the revised questions separate so you do not confuse a different test with an improved result.
Turn the result into a useful next step
The first check should answer a straightforward question: does this AI assistant suggest your business for these customer needs?
If it does, look at whether the recommendation is accurate and relevant. If it does not, examine the answers and sources, review how clearly your offer is presented, and decide whether broader testing would be worthwhile. If the results vary, keep that variation in your notes rather than choosing the screenshot you like best.
Return to the same questions after making meaningful changes, and compare the complete answers. A later appearance is worth recording, but it cannot by itself prove that a particular edit caused it.
Start with the service you most want to be hired for. Ask for a supplier as a customer would. Leave your name out of the question. That is how you begin to find out whether AI is introducing your business—or whether it only talks about you once someone already knows who you are.
If you would like this work handled for you, Lunasol’s GEO and AEO service includes AI visibility audits, entity work and content structured for AI citation.
AI Visibility is owned by Lunasol. The video-production examples in this article are illustrative.
