Customers ask AI and get the names of specific businesses
When someone today needs a repair service, a tradesperson, or an accountant, they don’t have to click through ten links. They type into ChatGPT or Copilot, “I need my laptop fixed, I’m in Olomouc, who should I contact” — and get three to five names with a short justification. Nobody approves those lists, and you can’t pay your way onto one. And you probably don’t know whether you’re on it.
That’s a new situation. Whether a customer even finds out about your business is now partly decided by a language model — a program that assembles answers from the texts it was trained on and from what it can look up on the web. The good news: you can check what it says about you yourself, for free, in ten minutes. The bad news: you might not like the result, and fixing it isn’t a single action but ongoing work.
A ten-minute test: ask like a customer
You don’t need any tool. Open ChatGPT (the free version is enough) and Microsoft’s Copilot, and ask the questions your customer would ask:
- “Recommend a [your field] in [your city].”
- “I’m looking for [specific service] in [city], who would you pick and why?”
- “Compare businesses that do [service] in [city].”
Try each question twice. Once with search turned on — ChatGPT then looks up current information on the web and shows source links with the answer. And once without it — the model answers only from what it “remembers” from the texts it was trained on. These are two different views of your business, and both count: the first shows how discoverable you are right now, the second shows how strong a trace you’ve left over the past years.
Vary the phrasing. Customers don’t ask uniformly — one types the name of the field, another describes the problem (“my computer won’t turn on, who can fix it in Brno”), another wants a direct comparison. Each phrasing can return a different list of names, which is exactly why it’s worth trying at least three variants, not just one.
For each attempt, note the date, the tool and model version used, the exact wording of the question, and the result: who appeared in the answer, whether you’re among them, and exactly what AI said about you. Finally, add a control question: “What do you know about the business [name] in [city]?” There you’ll see not only whether the model knows you, but also whether it isn’t saying nonsense about you — an old address, a wrong phone number, a service you don’t offer.
That record isn’t a formality. In a month you’ll ask the same questions again, and without notes you won’t be able to tell whether anything has changed.
Why AI knows some businesses and not others
We asked ourselves this exact question about a specific computer repair shop in Brno — and asked two versions of ChatGPT directly why they hadn’t recommended it. Both answered consistently, and their explanation is worth retelling, because it demolishes two widespread assumptions.
There’s no internal ranking of businesses. The model doesn’t have a table where businesses sit in ranked positions. Without search turned on, it picks based on how often and how consistently a brand appears across the texts it was trained on: in directories, review sites, articles, local listings. A business nobody has written about anywhere for years is practically invisible to the model — even if it’s the best in the city.
The business name won’t save you. An exact match between the name and the query isn’t decisive. “PC Service Brno” and the query “computer service Brno” are the same topic to the model; a descriptive name helps it understand what you do, but it doesn’t guarantee a recommendation.
With search turned on, the weighting shifts toward the present: what the search engine returns, what map profiles look like, how many reviews you have, whether the website clearly describes each service — and whether the same name, address, and phone number appear everywhere. The model cross-checks information from multiple independent places. When a directory lists a different address than the website, and a map profile lists a third one, there’s nothing to hold onto, and it reaches for a business it can be more confident about.
Bing is the gateway, not the elevator
ChatGPT’s search looks up websites mainly through the Bing index — an index being the list of pages a search engine knows about and can search through. If your website isn’t in Bing, ChatGPT simply can’t see it during search. Registering with Bing Webmaster Tools (the equivalent of Google’s Search Console, also free) is therefore a necessary condition that commonly gets overlooked here — attention usually stops at Google.
A necessary condition, though, isn’t a guarantee. OpenAI rearranges Bing’s results with its own system and supplements them with its own web crawling. Measurements show that Bing’s top three results match what ChatGPT actually cites only about 7% of the time. So ranking high in Bing alone isn’t enough — what matters is whether the business comes across as trustworthy and whether a clear answer to the person’s question can be pulled from its website.
In practice, watch two more things. In the robots.txt file — a text file on the website that tells search bots where they’re allowed to go — OpenAI’s bots must not be blocked: OAI-SearchBot, which crawls the site for search, and ChatGPT-User, which loads a page at the moment the model is assembling an answer from it. And expect a delay: changes on a regular website propagate into search-enabled answers within roughly one to three days, not instantly.
What a small business can do about it
None of the following requires an agency or a big-brand budget. It requires consistency:
- Create and maintain profiles. A business profile on Google, on local directories, and on maps — and Bing Places, the one most often forgotten, even though it’s exactly the profile layer of the search engine ChatGPT draws on.
- Keep your details consistent. The same name, address, phone number, and company ID on your website, in every profile, and in every directory. Every discrepancy lowers the confidence with which AI dares to recommend you.
- Collect real reviews. Continuously, from actual customers. Reviews are independent mentions the model can verify.
- Give each service its own page with a clear title in the format “Field City | Brand.” If you operate in multiple cities, each service should have a page for each location too. Service pages should also be followed by answers to frequently asked questions — how to write them is covered in FAQ on your website: how to answer customers and AI.
- Add a machine-readable business card. Schema.org LocalBusiness markup is a piece of code that tells bots, in black and white, what the business is called, where it’s located, and what it does. A visitor never sees it; a machine does.
- Build natural local mentions. An article in a local newsletter, a partnership, an industry directory — mentions beyond your own website are exactly the trace that lets the model recognize that you exist and can be trusted.
Most of this list is honest local SEO — covered in detail in Local SEO for a small business. Writing content so AI can quote it is a separate chapter, covered in How to write content so AI cites you. This article covers the step before that: finding out where you stand and understanding what you’re seeing.
Why you’ll get a different answer next time
You run the test, your business shows up in the answer, great. A month later you ask again and there’s nothing. Did something fail? Probably not — that’s just how AI answers behave. And the reverse is true too: if you don’t show up today, it doesn’t mean everything is wrong forever.
A different model version answers differently because it was trained on different data. The same version answers differently today than it will in a week, because answers aren’t a fixed list but a fresh composition every time. And the search-enabled mode reacts to whatever just changed on the website. That’s why a single measurement proves nothing — good or bad. Only one thing makes sense: ask repeatedly, with the same questions, log the results, and watch the trend over weeks and months.
From our own ongoing measurements, we have one sober observation: for customer-style questions, AI mainly recommends established, specialized brands with a long track record — mentions, reviews, years in existence. A new domain with a descriptive name starts from zero, and the name alone won’t pull it up. That’s not a reason to give up — it’s a reason to start building that trace now, because it will take months to grow.
When your own test is enough — and when you need measurement
The ten-minute test from the introduction is something every business owner can do, and should do today. It tells you whether you’re in the game, who’s in it instead of you, and whether AI is spreading outdated information about you.
What your own test won’t tell you is why. Whether you’re missing because your website isn’t in Bing, because of a mismatched address across three directories, because of blocked bots, or simply because not a single mention of the business has appeared outside its own website in ten years — that takes systematic measurement and diagnosis. That’s exactly what our AI Visibility Check is for: we repeatedly ask customer-style questions, record the answers over time, and deliver specific findings and a sequence of steps along with the results. If you want to measure it properly, the Check is part of our initial analysis — it reviews the website, search engine data, and real inquiries together, so you see AI visibility in context, not as an isolated number.
Whether you measure it or not, the question “what does ChatGPT say about us” has stopped being a curiosity. It’s a question customers ask on your behalf every day — and they’ll get an answer either way. Better to know it before they do.
