How does ChatGPT choose which businesses to recommend?

You ask ChatGPT for a good installer in your town and you see three competitors, but not yourself. Why? AI assistants do not work like a phone book. They put together an answer from what they know and what they find online at that moment.

Two sources: what the model knows and what it looks up

An AI assistant draws on two places:

• What the model learned during training. That is a snapshot of a large part of the internet. Businesses that are mentioned often and in many places are better represented in it.
• What the model looks up live. For questions about local businesses, ChatGPT, Gemini, Perplexity and Claude often search the web. They read a handful of pages and summarise them. Which pages depends on the search technology the model uses.

For local questions, the second part matters most. The good news: you can influence it.

Which pages do AI assistants read?

Look at the sources AI assistants cite for local questions and you see the same kinds of sites again and again:

• Review and comparison sites such as Google Maps, Trustpilot or industry-specific platforms.
• Lists and articles like "the 10 best hairdressers in Rotterdam" on local news or blog sites.
• Industry and business directories.
• The businesses' own websites, especially when they clearly state what they do and where.

If you are not on those sources, or only with thin information, the model has little to work with.

Which signals count?

Nobody outside the AI companies knows exactly how the choice is made. But measurements show clear patterns:

1. Relevance: does what you do match the question? A "plumber with emergency service" is mentioned more often for emergency questions when that literally appears on its site and profiles.
2. Location: is it clear which towns and neighbourhoods you serve?
3. Reputation: how many reviews do you have, how recent are they and what do they say?
4. Confirmation by multiple sources: when three different sites say the same thing about you, a model is more confident.
5. Readability: can the model open and understand your website? A site that blocks AI crawlers or puts everything in images is hard to read.

Why it differs per model

ChatGPT, Gemini, Perplexity and Claude do not use the same search engine or the same way of summarising. Gemini leans on Google, for example, while other models use other search indexes. So Perplexity may mention you while ChatGPT does not. That is why we always measure all four.

What can you do with this?

First look at which sources AI cites for questions in your industry and town. That is your biggest opportunity: if a model keeps using the same review site, you want to be well listed there. The report from the free AI check shows exactly which sources are cited and which competitors are mentioned in your area.