How AI Assistants Choose Which Local Businesses to Mention

Someone asks an assistant for a good roofer in their town. Three companies get named. You are not one of them, and one of the three has a website that looks like it was built in 2011.

That is the fact worth sitting with, because it tells you the answer was not built from web design.

Nobody publishes the local formula

Start here, because the market is full of people who imply otherwise.

None of the major engines publish how they select a local business for a recommendation. There is no local AI ranking factor list. Anyone showing you one built it from observation and called it a mechanism.

What is documented is narrower and more useful: what the retrieval step does, what the engines say matters generally, and what a page has to be for any of it to reach you.

A local question is not one question

Google names the technique behind its AI features: "Both AI Overviews and AI Mode may use a 'query fan-out' technique, issuing multiple related searches across subtopics and data sources, to develop a response" (Google Search Central, page last updated 2025-12-10, checked 2026-08-06).

Apply that to "best emergency plumber near me" and the shape changes. That one request becomes several: who covers this suburb, who works nights, what does a callout cost, who is licensed, what do recent reviews say.

Your homepage answers none of those. A page that names the suburb, states the hours in text, and gives a real price range answers three of them. Query fan-out covers the mechanics.

This is the practical reason local sites underperform their reputation. The business is genuinely good at the job and the site never states, in words a machine can lift, which areas it covers and what it costs.

Your website is one input, not the input

For a local answer, the engine has more to work with than your site, and often prefers it.

Directories, review platforms, local press, association listings, and map data all describe you, and they describe you in a consistent, comparable format that is easy to read. Your site describes you in a hero image and a form.

Google adds its own layer. Its AI features documentation lists, among practices that continue to be worthwhile, "checking that your Merchant Center and Business Profile information is up-to-date." That is Google naming its own business data as relevant to its own AI surfaces.

Note the boundary carefully. It is a statement about Google. It implies nothing about ChatGPT, Perplexity, or Claude, which have no access to your Business Profile and are working from the open web.

Which sets up the failure mode that catches almost everyone: your address on the site, your address in a directory from four years ago, and your hours on a third platform disagree. A human reconciles that in a second. A machine assembling a confident sentence has three candidate facts and no way to adjudicate, so it reaches for the business it can describe without hedging.

What the engines can and cannot see about you

Worth separating, because it changes where you spend an afternoon.

Google's AI features draw on Google Search, and Google holds business data about you independently of your website.

ChatGPT with search reads live pages through OpenAI's fetchers. Perplexity indexes the web with its own crawler and ranks against that index. Claude retrieves through its own search. None of them has a business profile to consult. For those three, the open web is the whole picture, and your site is a large part of the open web about you.

That asymmetry is the reason a single "AI local SEO" checklist does not exist. The engine that knows most about you is the one you can influence least directly, and the engines you can influence with a page are the ones with the least data about you.

The part you control, stated concretely

Four things, all checkable this afternoon, all on pages you own.

Services named in plain words. "Full-service solutions for your home" is unreadable. "Emergency drain clearing, water heater replacement, and leak detection" is three facts a machine can match against three different sub-questions.

Areas served as text. Not implied by a map embed, not left to the reader's inference from your address. Written out, on a page, in a sentence. How to structure service-area pages for AI answers covers doing that without turning it into a template farm.

Contact and hours as real content. Phone numbers inside images, hours rendered by a booking widget after load, and addresses that only exist in a footer script are all invisible to a fetcher reading raw HTML.

Markup that agrees with the page. Google's own list includes "making sure your structured data matches the visible text on the page." For a local business, Google's structured data documentation requires name and address, with properties like telephone, openingHoursSpecification, geo, and priceRange as recommended additions (Google local business structured data, page last updated 2025-12-10, checked 2026-08-06). Structured data for local business has the implementation.

Old way, new way

The old way was to rank in the map pack and to treat the website as a brochure the click landed on.

The new way is that a machine is composing a recommendation from several descriptions of your business, most of them written by somebody else, and your site is the one description you can rewrite. Its job changed from persuading a visitor to also supplying facts a parser can lift without guessing.

Both jobs still exist. The second one is new, and almost nobody has done it.

Why this drifts

Your hours change for a season and three platforms keep the old ones. You add a service and the site says so while the directories do not. A competitor gets written up in the local paper and becomes easier to describe than you.

None of that touches your website, and all of it changes what an engine has to work with. That is why readiness is something you maintain rather than a task you close, and the loop behind it is described on the how it works page.

The damaging admission

Here is what Citedon cannot see, stated plainly because this is the page where the limit bites hardest.

Our scan reads your website. It does not read your Google Business Profile, your directory listings, your review platforms, or map data. If your problem is that four sources disagree about your address, our scan will not surface that, and we are not going to imply it will.

What it does show is which of ChatGPT, Perplexity, Gemini, and Claude return or name your pages for the prompts a local buyer would actually type, and which competitor was named instead. That second part is usually the more useful half for a local business, because it turns an abstract worry into a name you recognize.

We cannot get you recommended. Selection is the engine's, it is probabilistic, and it changes. We measure whether your pages can be read and named, and we keep measuring.

And a genuine one about fit: if you are a single-location business with four pages, a phone number, and a booking form, the highest-value work here is an hour of writing your services and areas in plain text, not a subscription. The automated fix layer is WordPress only in any case, so on Squarespace, Wix, or a custom build you get the diagnosis and apply it yourself.

Where to start

Open the page a buyer would land on and read only what a machine gets: which services are named, which areas are stated, whether the hours and phone exist as text.

Run a free scan on that URL to see how the four engines read it today, and who they name instead. Then take the vertical version of this: local businesses, home services, dentists, medical practices, law firms, or real estate agents. The general selection mechanics, beyond the local case, are in how AI engines decide what to recommend.