Blog · How AI answers work

How AI assistants decide which local businesses to recommend

What OpenAI, Perplexity and Google say about where their answers come from, what that means for a local business, and the parts nobody outside those companies knows.

· 7 min read

Ask ChatGPT for "a good emergency plumber in Leeds" and it will usually give you three to six names, often with a sentence about each and a few links. If you run a plumbing business in Leeds, the obvious question is: why those names and not mine?

The honest answer has two halves. Some of it is documented by the companies that build these assistants, and you can act on it. Some of it is not documented anywhere, and anyone who claims to know it precisely is guessing. This article separates the two.

What the vendors say about where answers come from

All of the assistants your customers are likely to use now search the web before answering a question about local businesses, and they show the pages they used.

  • ChatGPT. OpenAI runs a crawler called OAI-SearchBot whose stated job is to surface websites in ChatGPT's search features. OpenAI's documentation for site owners (external site) says that sites which block it will not be shown in ChatGPT search answers. When developers use the same search through OpenAI's API, the web search guide (external site) describes inline citations for the pages used, plus a longer list of every page the model consulted.
  • Perplexity. Perplexity's help centre describes the product as searching the internet in real time (external site) and attaching numbered citations to every answer. Its crawler, PerplexityBot, is documented (external site) as the bot that surfaces and links websites in Perplexity's results.
  • Google AI Overviews and AI Mode. Google's guide to AI features for site owners (external site) says a page must be indexed and eligible to show in Google Search with a snippet to appear as a supporting link. It also says both features may use "query fan-out": running several related searches across subtopics and data sources to build one answer.
  • Claude. Anthropic lists a separate crawler (external site), Claude-SearchBot, for indexing content to improve search results it shows to users.

So the first, plain fact is this: an AI assistant mostly recommends local businesses that appear in the pages it finds when it searches. It is not consulting a secret list of approved plumbers. It is reading the web, quickly, and summarising.

The three kinds of page that tend to get read

When you look at the citations under answers about local services, the pages fall into three groups. This is our observation from reading answers, not a vendor statement, and the mix varies by engine and by trade.

1. Listing and review sites. Directories such as Yelp, Checkatrade, Tripadvisor, Houzz or Zocdoc, depending on the trade and the country. These pages are built to answer "who are the plumbers in Leeds?", so they match the question well, and they carry ratings, which an assistant can summarise in a sentence.

2. "Best of" articles and local roundups. A local newspaper's "ten best cafés in Bristol", a blog post, a trade association's member directory. These are lists written by a third party, which is exactly the shape of answer the assistant is producing.

3. Your own website. Your homepage, your services page, your contact page. An assistant can only describe what you do, where you work and when you're open if that information is written somewhere it can read. If your site says "we do everything" with no town names, it gives the assistant very little to match against "boiler repair in Headingley".

For Google's two AI features there is a fourth input that the others don't have in the same way: Google's own local data. Google's Business Profile help (external site) says local results are based mainly on relevance, distance and prominence, and that more reviews and positive ratings can help a business's local ranking. Google has not published how AI Overviews or AI Mode use Business Profile data, but its May 2025 guidance on AI search (external site) does ask businesses to keep their Business Profile information up to date. That is about as direct a hint as Google gives.

What this means in practice

If the assistants mostly repeat what the web says about your trade in your town, then your job is to make sure the web says something about you, in the places they read, and says it consistently.

  • Be on the listing sites that get cited for your trade in your area. Not every directory: the ones that actually show up under answers to your customers' questions. That list differs between a dentist in Manchester and a roofer in Ohio.
  • Make the basic facts identical everywhere. Your business name, address, phone number and hours should match across your website, your Google Business Profile and the main listings. If one site says you close at 5 and another says 6, an assistant has to pick one or leave you out.
  • Say plainly, on your own site, what you do and where. A services page that names the services and the towns you cover is easier to match to a question than a slogan.
  • Make your site readable by the search crawlers. If your robots.txt file blocks OAI-SearchBot or PerplexityBot, OpenAI and Perplexity say they will not show your site in search answers. We cover this in Should you block AI crawlers?.
  • Describe your business in a machine-readable way. LocalBusiness structured data won't guarantee anything (Google says there is no special schema needed for its AI features), but it states your name, address, phone and hours in a form no reader can misunderstand. See LocalBusiness schema in plain English.

None of this is new or clever. It's the same groundwork that helps in ordinary local search. What's different is that the assistant compresses ten results into three names, so the gap between being mentioned and being missing is much sharper.

What nobody outside these companies knows

It is worth being clear about the limits, because the market is full of confident claims.

  • How names are chosen from the pages read. No vendor has published how its model decides which three businesses to name out of the twenty it saw. Position in the source pages, review counts, how often a name appears across sources: any of these could matter, and the weighting is not public.
  • How much reviews count in AI answers specifically. Google says reviews help local ranking. It does not say how AI Overviews or AI Mode weigh them. The other vendors say nothing on reviews at all.
  • How stable any of it is. These products change their models and their search settings often, usually without notice. Something that worked last quarter may not work next quarter.
  • What any one person sees. Answers vary from one ask to the next, and the consumer apps can personalise results. We explain this in Why ChatGPT tells you something different from what your customers see.

Anyone selling a guaranteed way to "rank number one in ChatGPT" is selling something the vendors themselves do not offer. What you can do is put accurate information in the places these systems read, then measure whether you are named more often over time.

How Saymeter measures this

Saymeter asks the questions your customers ask across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and Google AI Mode, as a new customer in your town would, several times each week. It records every business each answer names and every site each answer cites, so you can see which listing sites matter for your trade and whether you're on them. The full method, including what we can't see, is on our methodology page. You can also run a free scan against three local competitors to see where you stand today.