How ChatGPT picks the local business it recommends
You type "recommend a bakery" into ChatGPT, with no city and no further explanation, and a moment later you have a list of businesses with stars, addresses and a sentence about each. But the answer doesn't come from one source. It is put together from a guess of your location, search results and business data from an internal ChatGPT tool. We show it step by step with bakeries in Žilina.
Five steps from question to answer
The user is shown only the finished result. We pieced the steps before it together from what independent research shows, as well as from the internal research of citeme.eu. We picked one concrete example for illustration. We go through each step separately below.
Five steps from the question "recommend a bakery" to ChatGPT's answer, with an example from Žilina
- Guessing where you are
If you don't write a city, ChatGPT uses an approximate location based on your IP address (it can be wrong), or what it remembers about you.
location guess: Žilina - Searching the web
ChatGPT turns the question into one or more search queries and gets results: mostly directories and a few company websites.
best bakery Žilina bakery reviews - Looking up business data
Besides the web search, ChatGPT may use something like a business search: an internal OpenAI tool that returns structured data directly, for example name, address and rating.
e.g. name, address, rating … for several businesses - Reading selected pages
From the results it opens the ones that look useful. About the rest it only knows what is in the snippet.
zlatafirma.eu · orlypekarstva.eu - Writing the answer
From the question, the page text and the business data the model generates the answer. The order and descriptions are its own.
1. Straff Bakery · 2. Crosta Bakery · …
The number of search queries depends on the model; some search more than once.
1. Guessing where you are
A customer can try it themselves: in an anonymous window, ask ChatGPT "guess where I am". ChatGPT answers with the city or area it guesses and itself adds that this is not an exact location. Such a guess from the IP address is not reliable and can be entirely off.
That ChatGPT uses an approximate location based on the IP address for local questions is also confirmed by an article by Cloutly that cites OpenAI's help centre. You can allow your device's precise location, but it is off by default.
If memory is on, what ChatGPT knows about you from earlier conversations also enters the context, and that can be a stronger signal than the IP address. With a more demanding task it may ask you directly. We turned memory off. For the example below we assume the guess came out as Žilina; for a customer it will be whatever it comes out as for them.
2. Searching the web
From the question the chatbot builds search queries (so-called fan-out queries), one or several depending on the model, and adds the city. In our example they were queries similar to these: "best bakery Žilina bakery reviews" and "bakery Žilina official offer". So it isn't the customer's question but a rewrite of it with the city, and a business has to be findable under such queries.
Of the 12 sites the chatbot found in this check, 6 were directories, listings and ordering portals, three were bakery and shop websites, two were baked-goods producers' websites and one was a job portal. Which directories and sources AI cites in different industries is covered in the article Where is it worth being for AI to find you.
The sites ChatGPT found when looking for a bakery in Žilina, by what they are (12 sites)
Blue marks directories where, according to their pages, a business can request verification or claim a profile. For Orly Pekárstva we found no such option.
* We couldn't verify the OMA site.
3. Looking up business data
Besides the web search, ChatGPT may use something like a business search. It is an internal OpenAI tool that returns structured data directly: business name, address, rating and the like. It may contain results from Google Maps, Yelp and other sources.
We don't know where exactly this data comes from. A study by Nicolas Sitter of 25,645 hotel queries and 99,538 businesses on map cards (US, Spain, Germany, UK; December 2025 to March 2026) found that most of the data came from Google Places: 88.8% over the whole period, already around 70.3% by the end of March. Research by Peec from September 2026 lists several routes through which Google Maps data reaches ChatGPT, and Yelp and Tripadvisor as licensed partners. For Slovakia, Google Maps is therefore the most likely source. That is a guess, not a confirmation, and our check supports it.
The bakeries from ChatGPT's answer on Google Maps and in the Orly Pekárstva and Zlatá Firma directories for Žilina
An illustrative result of one answer: it can differ when the question is repeated. The review count is in brackets. ChatGPT, Google Maps and Zlatá Firma give a rating out of 5, Orly Pekárstva out of 10. The order on Google Maps is from the search "pekaren zilina" and depends on where you search from. A dash means we did not find the bakery in the Zlatá Firma list we looked at.
The ratings in ChatGPT's answer match Google Maps exactly for all five bakeries, and the review counts differ by four at most. For both bakeries we could compare, Zlatá Firma has fewer reviews than ChatGPT's answer: Crosta Bakery 94 against 104, Pekáreň na Rohu 43 against 80. It states itself that it is based on Google Maps, and it shows a scan date of April 7, 2026, so it is an older copy of what Google Maps shows today. That suggests that, for Slovakia, the business data comes from Google Maps. Other sources such as Yelp or Foursquare aren't very widespread in Slovakia yet and have few reviews here; abroad it may look different.
4. Reading selected pages
From the results the chatbot first sees only short snippets, not whole pages, and then opens some of them. Which pages it opens usually isn't visible from outside; you can guess from the answer and sometimes find out exactly, for example in ChatGPT's Work mode or via the OpenAI API playground. In our example they were (probably) zlatafirma.eu and two profiles from orlypekarstva.eu: Straff Bakery and Pekáreň Alea. That is why it matters to be among the pages the chatbot finds.
5. Writing the answer
The last step is text generation: the model gets the question, the content of the pages it read and the business data, and generates the answer. In our example it recommended five bakeries: Straff Bakery, Crosta Bakery, Pekáreň na Rohu, Pekáreň Alea and Post Bakery. This is an illustrative result: we asked the question several times, and on repeating, the order and some of the bakeries differed (the second time, for example, kvap_ka appeared in the list), so results can always differ. The order in the figure above is not sorted by stars (second-placed Crosta has 4.3, fourth-placed Alea 4.8) and doesn't copy another source either: Orly Pekárstva puts Alea second, and Crosta isn't in its top three; Google Maps, for the search "pekaren zilina", shows Alea first and Pekáreň na Rohu only fifth.
What to do about it
- If you have a physical location, start with your Google Maps profile. Name, address, category, opening hours and replies to reviews. The stars and review count ChatGPT shows most likely come from there. This applies only to businesses with a physical location. In other countries, also look after your profile on Yelp, Foursquare and similar platforms: research lists them among ChatGPT's business data sources. We don't see their influence in Slovakia so far, but abroad it may matter. New: a business can now also create a business profile directly in ChatGPT (Businesses in ChatGPT). The business has to be verified.
- Find out which directories the chatbot reads. In citeme.eu you see this clearly in one place in Cited sources. Then correct your business details in those directories; in 4 of the 6 directories from our example a profile could be claimed or verified.
- Follow the AI search queries (fan-out queries). They show what the chatbot searches for when looking for your business and your competitors, so you know which queries you should appear under. In citeme.eu you find them in AI search queries.
- Follow your Google Maps reviews. Reply to them and take care that they are as good as they can be. The rating and review count are the numbers ChatGPT shows for a business. In citeme.eu this is included in the price: you find it in Profiles and reviews.
- Repeat. The answer changes from question to question and from person to person, one check isn't enough, and different chatbots read different pages.
citeme.eu does this for you: Local Visibility shows your business's position on Google Maps and the businesses ahead of you, Profiles and reviews your Google Maps profile and customer reviews, and AI visibility how often ChatGPT and other AI chatbots mention you.
How we looked into it
This is a check done by hand on November 18, 2026, which we ran several times; the answers differed (the order and some of the bakeries), so the article shows one of them as an illustration: ChatGPT with memory off, the question "recommend a bakery" with no city, location Žilina. It is one city, one type of business, one chatbot and one day. We read the ratings and review counts by hand from ChatGPT's answer, from Google Maps, from Orly Pekárstva and from the Zlatá Firma page (its data carries a scan date of April 7, 2026). The list of sites and search queries comes from the same check.
Frequently asked questions
How does ChatGPT know where I am?
From an approximate location based on the IP address that the app or browser sends it, and if memory is on, from earlier conversations. It uses your device's precise location only with your permission. The guess may not be accurate; if you want certainty, put the city in the question.
Where does ChatGPT get the stars and review count for businesses?
From an internal business-data tool that ChatGPT may use alongside the web search. We don't know exactly where it draws from. Research shows that for hotels most of the data comes from Google Places, with Yelp and Tripadvisor appearing alongside. In our check the ratings in ChatGPT's answer matched Google Maps exactly for all five bakeries and the review counts differed by four at most, while Zlatá Firma had fewer reviews.
Is a good website enough for ChatGPT to recommend me locally?
Not on its own. In our example half of the sites found were directories and listings, not bakery websites, and the rating in the answer matched Google Maps. So besides a website you need (if you have a physical location) an up-to-date Google Maps profile and correct details in the directories the chatbot reads.