How AI recommends local businesses

The local recommendation stack

The useful mental model is a stack. An assistant doesn’t hold its own database of every local business and a ranked opinion of each one. Instead it sits on top of other people’s data and reassembles it on demand. Picture four layers feeding the answer:

• A place layer— structured listings that say a business exists, where it is and what category it’s in (Foursquare, Google Business Profile, Yelp, Apple Maps).
• A reputation layer— reviews, ratings and the language inside them that say what the business is actually like.
• A corroboration layer— mentions and citations on other pages the engine already trusts, which say “other sources agree this business is real.”
• A self-description layer— the business’s own website and schema, the one source it fully controls, which says “here, in plain text, is exactly who we are.”

Each layer has a different owner and a different weight. The important consequence: the layer you control most directly (your website) is usually notthe one that carries the most weight for local recommendations — the structured listings do. That’s why owners who pour effort into their site but ignore their listings often stay invisible in AI answers.

An assistant doesn’t decide who’s best. It surfaces whoever the place, reputation, corroboration and self-description layers already agree on.

The signals, ranked by weight

Here are the inputs that feed those layers, ordered roughly by how much they tend to move local AI recommendations. Engines weight them differently and the details shift over time, so treat the order as a working guide rather than a fixed formula.

1. Structured local listings (the biggest lever)

   Map and directory databases — Foursquare, Google Business Profile, Yelp, Apple Maps — are the cleanest machine-readable record of where a business is, what category it’s in and whether it’s open. Assistants lean on these heavily for “near me” answers. For ChatGPT specifically, its local results reportedly draw on a combination of Foursquare and Google Business Profile data — on the order of 70%of that local data between them — with Foursquare reportedly the larger of the two. You can’t edit Foursquare as directly, so an accurate, complete Google Business Profile is the practical lever most owners can actually pull.

2. Reviews and ratings

   Volume, recency, average rating and the actual words inside reviews all feed the picture. An assistant asked for the 'best' option leans on corroborated sentiment, and review text is a rich source of the specifics ("great for kids", "open late", "vegan options") it quotes back. Reviews you didn't write about yourself are treated as more trustworthy than your own marketing copy.

3. NAP consistency across the web

   Your name, address and phone number(NAP) should read identically everywhere they appear — your site, your listings, the directories that mention you. When they conflict (an old suite number, a tracking phone line, “Co.” in one place and “Company” in another), the engine can’t confidently merge those records into one business, so it hesitates to name you. A quick check: search your exact business name and confirm the top results all show the same address and phone.

4. Citations and mentions across the web

   Being named on pages an engine already trusts — local press, reputable roundups, chamber and association directories, supplier pages — is corroboration. The more independent sources describe your business consistently, the more 'real' and recommendable it looks. A business that exists only on its own website has nothing backing up its claims.

5. Your own site's content and entity signals

   This is the one source you fully control. Pages that state in plain HTML who you are, where you operate, what you offer and what it costs — reinforced with LocalBusinessschema — give the engine crawlable, quotable facts. View Source on your own page: if your name, address and service area aren’t in the raw HTML, the AI layer can’t read them either (here’s how to test what AI can see).

Why Foursquare keeps coming up

Owners are often surprised that Foursquare — a name many associate with check-ins a decade ago — matters so much. The reason is plumbing: Foursquare maintained one of the largest structured databases of physical places, and that data is licensed into products that need a clean “places” dataset, reportedly including ChatGPT’s local results.

How to read your own stack

You don’t need special tools to audit most of this — you can check each layer by hand:

• Place layer:search your exact business name in Google and on Apple/Google Maps. Do your name, address, category and hours match across every result? Mismatches are the most common reason an engine can’t merge your records.
• Reputation layer: count your reviews and read the most recent ones. Are they recent, plausible and specific enough to quote? A profile with three reviews from two years ago gives an assistant little to work with.
• Corroboration layer: search your business name in quotes. How many independent, reputable pages name you, and do they agree on the details?
• Self-description layer: open your own site, right-click and choose View Source. If your name, address and what you do aren’t in that raw HTML, the AI layer can’t read them either.

Frequently asked questions

Keep going