How it works
How do ChatGPT and other AI apps choose which businesses to name?
By Last updated
When an AI app like ChatGPT is asked to recommend a business, it tends to search the web, read a small number of pages, and write an answer from what those pages say. It can only name what it found, so businesses the sources describe clearly and mention often are easier to name. None of the AI companies publish their full method, but their documentation, and how answers are built, show enough of the pattern to act on.
This guide explains each step in plain English, what is known for certain, and what is not.
What happens when someone asks ChatGPT for a recommendation?
For a question like "who is a good emergency plumber in Croydon", a modern AI app goes through roughly three stages.
- It decides whether to search. Questions about local businesses, prices or recent products are a natural fit for a web search, since the AI's built-in knowledge can be out of date and lacks current reviews or opening hours. Google, for example, builds its AI features on Google Search (Google Search Central).
- It searches and reads. The app runs one or more searches, picks a handful of results, and reads them. Google describes its own AI Mode and AI Overviews as using "query fan-out", which means "issuing multiple related searches across subtopics and data sources" to build a response (Google Search Central).
- It writes the answer. The AI summarises what it read, picks the businesses it thinks fit, and often lists the websites it used. Anthropic says every Claude answer that uses web search "includes citations, so you can easily verify sources yourself" (Anthropic).
The important point for an owner: in stage 3, the AI can only name what it found in stage 2.
Where does ChatGPT get its information about businesses?
From two places: what the model learned during training, and what it finds when it searches. For recommendations, search matters most.
AI companies run their own programs, called crawlers or bots, that visit websites. Each company separates the bot used for search from the bot used for training:
| Company | Search bot (affects answers) | Training bot | What the company says |
|---|---|---|---|
| OpenAI (ChatGPT) | OAI-SearchBot | GPTBot | Sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers" |
| Anthropic (Claude) | Claude-SearchBot | ClaudeBot | Blocking the search bot "may reduce your site's visibility and accuracy in user search results" |
| Perplexity | PerplexityBot | (not used for training) | PerplexityBot is "designed to surface and link websites in search results on Perplexity" |
| Google (AI Overviews, AI Mode) | Googlebot | Separate controls | AI features use normal Google Search; "no additional requirements to appear" |
Sources: OpenAI, Anthropic, Perplexity, Google.
This matters because many owners block "AI bots" to stop their content being used for training, and accidentally block the search bot too. Blocking a training bot and blocking a search bot are two different decisions.
Why do AI apps prefer review sites and lists over business websites?
Because a recommendation is a comparison, and comparisons live on other people's websites. A business's own site says it is great; a review site, a "best of" article or a directory compares several businesses side by side. That is exactly the material an AI needs to write "here are four good options".
Namova's own check shows how strong this pull can be. On 4 Nov 2026, Namova asked three AI models, through their APIs, 10 mattress questions, each asked twice, for 60 answers. One review site was cited in 50 of those 60 answers. Two brands were named in about 2 in 3 answers each (42 and 40 of 60), while most other brands appeared far less often.
The lesson is not "give up on your own website". It is that your website is where facts get checked, while other websites are where you get compared. You need both.
Why does ChatGPT give a different answer every time?
AI answers are generated fresh each time, and small differences in wording, search results and the model's own randomness change which businesses make the list. In Namova's mattress check, when the same question was asked twice on the same AI model, it named exactly the same set of brands in only 3 of 30 repeat pairs.
Other things also change the answer:
- The app and model. ChatGPT, Claude, Gemini, Perplexity and Grok search differently and write differently.
- Location and account. The consumer apps can use your location, memory and past chats. The versions developers use through an API usually do not.
- Time. Search results and the AI models themselves are updated often.
So a single screenshot is weak evidence. If an agency or a tool shows you one answer and calls it your "AI ranking", ask how many times the question was asked, on which apps, and what the range was.
What makes a business easier for AI apps to name?
No AI company publishes a scoring formula. What is safe to say, from their documentation and from how answers are built, is that a business is easier to name when:
- Its ChatGPT business profile is claimed, where eligible. Since 10 November 2026, OpenAI's Business Hub (in beta) lets businesses claim and verify a profile in ChatGPT, which lets them manage their details; OpenAI says claiming does not guarantee better placement (OpenAI, VentureBeat).
- Its website can be read by the search bots above, and indexed by Google.
- Its facts are plain and consistent: what it does, where, prices or price range, hours and contact details, the same everywhere.
- Other sites mention it: reviews, directories, comparison articles, local news, industry lists.
- Its pages answer real questions directly, so a sentence can be lifted into an answer without losing its meaning.
Research supports the last point. A 2024 study by researchers at Princeton and elsewhere, presented at KDD, tested changes like adding sources, quotations and statistics to web pages and found these methods "can boost visibility by up to 40%" in the answers of the generative engines they tested (Aggarwal et al., arXiv). That was a controlled test on the authors' own benchmark, not a promise for any business, but it shows that how a page is written can change whether it gets used.
For the practical steps, see how to get your business recommended by ChatGPT.
Do reviews and Google Business Profile affect AI answers?
For Google's own AI answers, very likely, since they are built on Google Search; for other AI apps, no company says how much. Google's own local results weigh relevance, distance and prominence, and Google says "more reviews and positive ratings can help your business's local ranking" (Google). Review sites and map listings are where local businesses get compared, so they are likely sources for other AI apps as well.
The guide to Google reviews and Business Profile for AI answers covers this in detail.
What do we not know?
It is worth being honest about the limits:
- No AI company publishes how it ranks the businesses it names.
- The consumer apps can show different answers to different people, and nobody outside the company can see every person's screen.
- A change in your numbers after you fix something does not prove the fix caused it. The AI app may have changed too.
That is why Namova reports results as ranges, asks each question more than once, and labels every AI app. It measures the pattern across many answers; it does not claim to know what any one customer is shown. To see the pattern for your own business, start with a website snapshot or look at the plans and prices.
Frequently asked questions
Does ChatGPT recommend businesses that pay?
No fee buys a place in the normal ChatGPT answer. Where an AI app sells advertising, it is a separate product from the answer itself.
Can I block AI apps from training on my site but still appear in answers?
Yes, in most cases. OpenAI and Anthropic each use a separate bot for search and for training, so you can block the training bot in robots.txt and still allow the search bot. Perplexity says its search bot, PerplexityBot, is not used for training at all. Check each company's documentation for the exact names.
Why does ChatGPT name my competitor and not me?
Usually because your competitor is better covered on the websites ChatGPT read for that question: more reviews, more listings, or a mention in a comparison article. Look at the sources cited in the answer to see where the gap is.
Does it matter which AI app my customers use?
Yes. Each app searches differently and can name different businesses for the same question, which is why it is worth checking more than one.
How often do AI answers change?
Often enough that one check is not a measurement. Ask the same questions more than once (2 to 3 times), on more than one app, and compare ranges over weeks.
Sources
Every link was checked on 14 November 2026. Namova is independent and not affiliated with any company named here.
- Businesses in ChatGPT announcement, OpenAI, 10 Nov 2026
- AI features and your website, Google Search Central
- Overview of OpenAI crawlers, OpenAI
- Does Anthropic crawl data from the web?, Anthropic
- Enabling and using web search, Anthropic
- Perplexity crawlers, Perplexity
- Tips to improve your local ranking on Google, Google
- GEO: Generative Engine Optimization, Aggarwal et al., KDD 2024
- Namova check, 4 Nov 2026: 10 mattress questions on OpenAI GPT-5.4 mini, Perplexity Sonar and Anthropic Claude Haiku 5.5, each through its API with web search on, each asked twice (60 answers). Brand and site names hidden.