How AI assistants pick which local businesses to mention

The search behind the answer

People increasingly skip the list of blue links and ask a question instead. "Who is a good electrician near Clermont" typed into ChatGPT, Claude, Perplexity, or Google's AI results returns a short written answer with a handful of business names in it. If your business is one of the names, you got a referral that cost you nothing. If it is not, the customer may never see the search results page where you rank sixth.

Under the hood these assistants do something recognizable. They run searches, read the pages that come back, cross-check what they find against listings and reviews, and compose an answer from businesses they can describe with confidence. That last phrase is the whole game. An assistant will not name a business it cannot confidently describe, because a wrong recommendation makes the assistant look bad. Your job is to be easy to describe.

Signal one is structured data

Structured data, usually called schema markup, is a block of code on your website that states the facts about your business in a format machines read directly: what you are called, what trade you are in, your address, your hours, your service area, your phone number, your reviews. Humans never see it. Crawlers see nothing else so clearly.

A page without schema forces the machine to infer what you do from your paragraphs, and inference is where errors and omissions happen. A page with LocalBusiness schema, plus the more specific types like Roofer, Dentist, or Plumber where they exist, hands over the facts in a form that cannot be misread. FAQ schema does the same for the questions customers ask, which is content assistants love to lift answers from.

Signal two is the llms.txt file, and honesty requires a caveat

robots.txt has told crawlers where they may and may not go since the nineties. Its newer sibling, llms.txt, is a plain text file written specifically for AI crawlers that describes what your business is, what services you offer, and where you work, in clean prose the crawler can ingest without parsing your layout.

Here is the honest state of it in 2026: an Ahrefs study of over 137,000 sites found no evidence that any major AI company actually fetches the file yet. So we will not tell you llms.txt gets you recommended, because right now it does not. We include it anyway because it takes minutes to write, costs nothing to host, and if the crawlers ever do start reading it, yours is already there. Treat it as a checkbox, not a lever. The levers are the other four signals on this page: schema, readable reviews, one page per question, and consistent business data. Anyone selling llms.txt as the secret to AI visibility is selling the cheapest line item in the package as the headline.

Signal three is reviews the machine can read

Assistants lean hard on reviews because reviews are the closest thing they have to ground truth about whether a business is any good. They read your Google profile, Yelp, Facebook, and trade directories, and they also read your own website. Most business sites either show no reviews at all or render them through a third-party widget that loads with JavaScript, which many crawlers never execute. The review exists, the machine never sees it.

The fix is a reviews page in plain HTML: your real reviews, with dates and sources, as ordinary text on your own domain. Now the assistant reading your site finds not just a claim that you do good work but specific customers saying so in quotable sentences. To be clear about what this is not: it does not mean inventing reviews, which is both illegal and exactly the kind of thing these systems are getting better at spotting. It means making the real ones readable.

Signal four is a consistent name, address, and phone number

Assistants verify. Before naming you, the system effectively asks whether the business on your website is the same business on Google, on Yelp, and in the county directory. If your site says "Smith & Sons Roofing LLC" while Google says "Smith and Sons Roofing" at an old address with a disconnected number, that verification gets murky, and murky businesses get left out of answers in favor of ones whose facts line up everywhere.

The industry calls this NAP consistency, for name, address, phone. It is tedious work rather than clever work: list everywhere your business appears, and make every listing match, character for character. Old phone numbers and stale addresses from a move three years ago are the usual offenders.

Signal five is pages that answer one question each

A single homepage that mentions eleven services in passing gives an assistant very little to work with when the question is specific. A page dedicated to water heater replacement in Mount Dora gives it everything: a service, a place, and details worth quoting. This is why service pages and location pages have become the backbone of local visibility, in AI answers and in ordinary search alike.

The standard is one page per service you want to be found for, and one per town you work in, each with real content about that service in that place rather than the same paragraph with the town name swapped. Thin, duplicated pages are ignored by search engines and assistants both. A handful of substantial pages beats forty copies.

What this looks like as a to-do list

Add LocalBusiness and FAQ schema to your site, and the specific business type where one exists. Write and upload an llms.txt file. Put your best real reviews on a plain HTML page of your own. Make your name, address, and phone identical everywhere you are listed. Build a real page for each main service and each main town. None of it is exotic, and all of it can be done on the website you already have, whoever built it.

If you would rather hand it off, this exact list is our Get found package: $35 a month, installed on your existing site, no rebuild required. It is also included in the Diamond plan at $215 a month along with the rest of our tools. What no one can sell you, us included, is a guaranteed placement in an AI answer. The systems change, the answers vary by phrasing and day, and anyone promising a guaranteed mention is guessing on your dime. What you can buy is the state of being easy to verify and easy to quote, and that is the whole strategy.