How many businesses does ChatGPT recommend when a customer asks for one?
EasyFound · · 7 min read
Usually three to five. When a customer asks ChatGPT to recommend a business, it names a handful, not a long page of links. In our Malaysian tests, most answers stayed small, and the name at the top usually changed when the question was asked again. So aim to be in the handful often, not to rank first.
ChatGPT usually recommends three to five businesses when a customer asks for one. In our Malaysian tests, an answer that named businesses named four as the median, and about four in five named five or fewer. It's a handful, not a page of ten links.
Picture a customer typing "My aircon at home is not cold anymore. Can you recommend a good aircon servicing company in Johor Bahru?" On Google they'd scroll past ten links and a map. ChatGPT gives them a few names and stops. And the handful changes each time someone asks.
How many businesses does ChatGPT name in one answer?
ChatGPT named between three and five businesses in most of its recommendation answers in our tests. We counted the businesses in 377 saved ChatGPT answers to Malaysian buying questions, from clinics and IT support to nasi lemak. Of those, 324 named at least one business:
| Businesses named | Answers |
|---|---|
| 1 | 27 |
| 2 | 32 |
| 3 to 5 | 209 |
| 6 or 7 | 41 |
| 8 or more | 15 |
So fewer than one answer in ten named just one business, and the longest list had 11 names. The other 53 answers named no business at all. The biggest group was company secretary questions, where ChatGPT gave "how to choose" advice instead.
Does the number change by industry or by how the question is asked?
The kind of business changed the count more than the wording did. Questions from overseas buyers about electronics manufacturers drew the longest lists, level with hotel questions. Company secretary questions drew the shortest, and ChatGPT named a firm in only 6 of 27 of them.
| Kind of business | Typical number named |
|---|---|
| Electronics manufacturers (overseas buyers) | 6 |
| Hotels | 6 |
| WhatsApp chatbot builders | 5 |
| Nasi lemak places | 5 |
| Accounting and tax firms | 5 |
| Aesthetic and skin clinics | 4 |
| IT support companies | 4 |
| Renovation contractors | 4 |
| Aircon servicing | 3 |
| Freight forwarders | 3 |
| Solar installers | 3 |
| Tuition centres | 3 |
| Company secretaries | 1 or 2 |
How the question was asked mattered less:
- Short "best X in KL" questions vs detailed ones. Both had a median of four names. Short questions ran to six or more names in about one answer in five, detailed ones in about one in seven.
- Naming the area vs "near me". In our city test, naming KL, Penang or Johor Bahru instead of saying "near me" left the median at three.
Each row comes from one or two tests, with different questions and model versions, so read the table as a rough guide, not a league table.
Does ChatGPT name the same business first every time?
ChatGPT rarely kept the same business at the top when we asked again. Across 208 pairs of repeat answers to the same question, the first name matched in 63, about three in ten. In six repeats in ten, the business named first in one answer wasn't named anywhere in the other.
The exception was food with a few famous names. In our nasi lemak check, ChatGPT opened both runs with the same KL place for 7 of 8 questions. The service businesses in our tests rarely got that.
SparkToro's study across ChatGPT, Claude and Google's AI, mostly with US examples, saw list length swing too: as few as 2–3 recommendations, and "equally often 10 or more"1. Why a ChatGPT rank position isn't worth tracking is covered in how AI visibility tools measure ChatGPT. Why the list moves at all is covered in why ChatGPT gives a different answer every time.
How big is ChatGPT's shortlist pool for one question?
ChatGPT's pool for a question is bigger than any one answer. Two runs of the same question together named a median of seven different businesses, while each answer named about four. Short "best X" questions drew a wider pool than detailed ones, though most of that gap came from one test, so treat it as a hint.
That's the useful part for a business owner. A handful of slots sounds harsh, but the handful is redrawn from a wider pool each time. If pages about you keep turning up in ChatGPT's searches, you can be in the next draw, even if a competitor was named today.
What can't our tests tell me about how many businesses ChatGPT names?
- The ChatGPT app. We used OpenAI's API with web search. The app, with maps, memory and your location, may show more or fewer names.
- Long-run steadiness. Each question was asked only two or three times, between late September and early October 2026.
- Exact industry gaps. Different tests used different questions and model versions, so small gaps between kinds of business may be noise.
How do I get my business into ChatGPT's handful?
Our reading: few names were locked in at the top in our tests, so a business ChatGPT can read clearly has a chance in each new draw.
- Request EasyFound's free check. See what ChatGPT says when customers ask for a business like yours, then go through it with us on a free call.
- Say what you do and where, in plain text. Those are the two facts ChatGPT repeated most when it named a business, as our test of what ChatGPT repeats shows.
- Make your website readable to ChatGPT. Run through the website readiness checklist, from crawler access to a page per service and area.
- Count how often you're named, not where. Ask your customers' real questions several times, logged out, and track the share of answers you appear in.
EasyFound is an AI SEO service for Malaysian businesses. We check how often you're named across six AI tools in English, Malay and Chinese, then work on the pages they draw on. The free check below is the first step.
Are you in ChatGPT's handful?
See what ChatGPT says when customers ask for a business like yours, then go through it with us on a free call.
Frequently asked questions
Does ChatGPT only recommend one business when I ask for one?
Rarely. In EasyFound's Malaysian tests through OpenAI's API, fewer than one in ten ChatGPT answers that recommended businesses named just one. The median was four. We also asked ChatGPT itself, four times. Half the runs said one main pick; the other half said 'usually 3 to 5'. Its real answers matched the second.
Is the first business ChatGPT names the best one?
Not in a way you can rely on. In EasyFound's Malaysian tests, when the same question was asked again, the business named first stayed first in only about three repeats in ten. In six repeats in ten, the business named first in one answer didn't appear anywhere in the other. The exception in those tests was food with a few famous names, such as nasi lemak in KL.
How many businesses does ChatGPT choose from for one question?
More than it shows in one answer. In EasyFound's Malaysian tests, two runs of the same question together named a median of seven different businesses, while a single answer named about four. Each answer is a fresh draw from a wider pool, which is why a business missing today can be named tomorrow.
Does ChatGPT name more businesses if I say my area or ask in more detail?
Not much, in EasyFound's Malaysian tests. Saying the city instead of 'near me' left the median at three businesses in the city test. Short 'best X in KL' questions and detailed questions both had a median of four, though short ones more often produced longer lists of six or more. The kind of business mattered more.
Sources
Our data: EasyFound, How many businesses ChatGPT names per recommendation answer, October 2026. We made no new AI calls for the counts: every count comes from 377 saved answers to 152 Malaysian buying questions in eight earlier tests, all through OpenAI's Responses API with web search and approximate location Malaysia, not the ChatGPT app. The tests: a language test of five trades in English, Malay and Chinese (24 September, gpt-5.4-mini, three runs, short keyword questions); a city test of aircon, renovation and tuition questions with and without a city (2 October, gpt-5.4-mini); 30 aesthetic and skin clinic questions (29 September, chat-latest); and IT support, WhatsApp chatbot, freight forwarder and electronics manufacturer checks (2–6 October) and a nasi lemak check (8 October), all chat-latest. Each answer's count comes from that test's published data, counted once per answer, so the language test matches our answer-stability data and clinic branches and spellings are merged. Which business came first, and the pool across runs, come from the businesses listed in the order each answer named them; for the language test that list is our earlier, unfiltered extraction, which can include directories. Filtering it the way our answer-stability data does left the first-name and pool figures unchanged. First-name and pool comparisons use every pair of runs of the same question; the freight and electronics checks recorded counts only, and the nasi lemak check recorded only the first place, so they're left out of the pool figure. "Detailed" questions are full sentences with context; "short" ones are keyword phrases such as "best managed IT service provider in Kuala Lumpur". The quoted self-description comes from four extra runs on 9 October asking ChatGPT how many businesses it recommends. Download the per-answer counts (CSV). We don't name the businesses, because a few test runs aren't a ranking.
Other sources:
- SparkToro (Rand Fishkin and Patrick O'Donnell), NEW Research: AIs are highly inconsistent when recommending brands or products, 28 January 2026; 2,961 responses from 600 volunteers, November–December 2025 (link)
The counts are shared under CC BY 4.0: you're free to reuse them, with credit to EasyFound.
ChatGPT changes often; treat these figures as a dated measurement.
Research by Janice Pang, EasyFound.