Someone's boiler fails on a Sunday evening. Instead of searching Google they open ChatGPT, type "who can fix a boiler in Bristol tonight", and get three names with a sentence of reasoning under each. None of those three businesses did anything special for AI. They were simply the ones the assistant could find, read and explain when it went looking.
That is the whole mechanism. An assistant has no private knowledge of your town. It looks you up in the same public sources Google uses, and most of those sources are things you already control.
An assistant doesn't know your town, it looks it up
A language model is trained on a snapshot of text that ends months before you use it, and that snapshot says little about a plumber in one English city. So for a local, current question the assistant runs searches and reads what comes back before it writes a word. Google documents this for its own AI features: its page on AI features and your website describes AI Mode and AI Overviews issuing several related searches for one question, which Google calls query fan-out, and composing an answer from the pages those searches return.
The other assistants work the same way in outline. OpenAI names the crawler that surfaces sites in ChatGPT search (OAI-SearchBot, separate from GPTBot, which it uses for training), and Perplexity names PerplexityBot and shows its sources beside every answer. What none of them publish is how they choose between two boiler engineers once the sources are in. Everything below about that choice is observed by asking assistants repeatedly, not read in documentation, and it can shift without notice.
The sources that come back
Ask assistants local questions across a few trades and towns and the same four kinds of source keep appearing, in roughly this order of weight.
- Business listing data. For Google's AI surfaces this is your Google Business Profile: name, category, address, hours, rating and review count, often shown as a place card in the answer. Other assistants read the same facts through Maps listings and directories.
- Review sites. Google reviews first, then whichever platforms matter for your trade: accreditation sites for builders, travel review sites for restaurants, professional directories for clinics. The assistant reads the text of reviews, not just the star count.
- "Best X in Y" lists. Local newspapers, city guides and bloggers who rank the ten best dentists in town. These pages are already shaped like an answer, so an assistant leans on them heavily and often repeats their wording.
- Your own website. Service pages that say what you do, where, for whom and on what terms. This fills the gaps the others leave: emergency call-outs, whether you treat children, Saturday opening.
Notice what is on that list. Your Profile, your reviews and your pages are yours to change this week. The lists are earned, and the local links chapter covers how.
How the shortlist gets made
With the sources gathered, the assistant has to pick a handful of names and justify them. Watching this across many questions, three patterns hold.
It favours businesses that appear in more than one source with the same details. A clinic on Maps, two review sites and a city guide, with the same name, address and phone everywhere, looks like a confirmed entity. One that appears once, or with a different phone number in each place, gets dropped or described vaguely.
It needs a reason it can say out loud. Assistants write "praised for same-day appointments" or "several reviews mention clear pricing" because those phrases exist in the review text or on a page. A business with a high rating but reviews that say only "great service" gives the assistant nothing specific, so it tends to name the competitor whose reviews carry detail.
It matches the question's wording to categories and services. "Emergency electrician" pulls in businesses whose Profile category, listed services or page headings use those words. The relevance rule Google states for local search in its Business Profile help is doing the work here, by a different route.
Why this overlaps with the local SEO you already do
The inputs are the ones you have been working on since the start of this track: a complete Profile, steady reviews, consistent listings, pages that rank. That is good news, because nothing here asks you to start a separate project.
The difference is in what the sources have to contain. Google's map ranking can place a profile on category, distance and review count alone. An assistant has to write sentences, so it needs text: review content with specifics, a page that states your terms, a description that says who you serve. A Profile with the right category and a phone number can rank in the map pack and still give an assistant nothing to say about you.
Location differs too. When the question names a town, that town drives the search. When it doesn't, each assistant guesses where the person is, which is one reason two people in the same street get different names. Treat any single answer as one sample; the last lesson in this chapter shows how to measure properly.
What to take away
- An assistant has no knowledge of your town from training; it searches live and reads what comes back, which Google documents as query fan-out for its own AI features.
- The sources it reads are your Business Profile, review sites, "best in town" lists and your own pages, and you control three of the four.
- How it chooses between businesses is observed rather than documented, but it consistently favours names confirmed across several sources with specific, quotable reasons.
- The local SEO work you already do supplies the inputs; the extra job is making sure those inputs contain sentences an assistant can use.
Next
Now that you know what assistants read, Being the business AI names puts the four fixes in the order that matters.