How ChatGPT Decides Which Local Businesses to Recommend
It isn't guessing and it isn't random. ChatGPT builds confidence from convergent evidence — and the businesses that understand the mechanism get named.
When ChatGPT tells someone "I'd recommend Riverside Dental — they have excellent reviews and specialize in implants," it isn't guessing, and it isn't pulling from a hidden ranking. There is a knowable mechanism behind which businesses get named. Understanding it is the difference between marketing to AI blindly and marketing to it deliberately.
Two brains: training data and live search
ChatGPT answers local questions with two different systems, and they reward different things.
The model's memory (training data). The base model was trained on a snapshot of the web. If your business has years of consistent mentions — directories, reviews, news, your own site — you may live in that memory. This is slow to change: it rewards businesses with a long, stable footprint.
Live web search. For anything current — "open now," "near me," prices — ChatGPT searches the web through Bing's index, reads the top results, and synthesizes an answer with citations. This is fast to change: it rewards businesses that rank in traditional search and publish quotable, factual pages.
The practical takeaway: your traditional SEO still matters enormously, because it feeds the live-search half. If you rank on Bing and Google for "emergency electrician springfield," you're in the pool ChatGPT draws from. If you don't, you aren't.
Why AI names two businesses, not ten
A search results page distributes attention across ten links; a conversation can't. Language models are built to give a confident, useful answer, so they compress everything they know into one or two named recommendations plus reasons. Internally, the model is looking for convergent evidence: multiple independent sources that agree this business exists, does this service, in this place, and is well regarded. The businesses that get named are the ones where every source tells the same story.
The signals that build that confidence
Entity consistency. Identical name, address, and phone across your site, Google Business Profile, Yelp, and directories. Mismatches fragment you into several "maybe" entities, none strong enough to recommend.
Review mass and recency. Reviews are the loudest third-party evidence of quality. Volume matters, recency matters more — a wall of 2022 reviews reads as "possibly closed."
Quotable pages. When live search fetches your site, the model extracts sentences that answer the question. Pages structured as direct answers — service, price range, area served, in the first paragraph — get quoted. Pages of marketing fluff get skipped.
Structured data. LocalBusiness and FAQPage schema hand the model your facts in its native format. It is the cheapest confidence you can buy.
Third-party corroboration. News mentions, association memberships, "best of" lists. One independent source saying "trusted local roofer" outweighs ten self-descriptions.
What this looks like in practice
Picture two competing HVAC companies. Company A has a five-year-old website that says "quality service you can trust," 40 reviews, and a Yelp listing under a slightly different name. Company B publishes "AC repair in Mesa costs $120–$380 for most jobs — here's what affects it," carries 300 reviews with 20 from last month, has identical listings everywhere, and marks it all up with schema. When someone asks ChatGPT who to call, Company B gets named. Not because it gamed anything — because it gave the model evidence it could verify from three directions.
Test it on your own business
Ask ChatGPT (with browsing) the question your customer would ask, in three phrasings. If you're not named, the fix is almost always in the signals above. Our AI Visibility Checker audits the on-site half automatically, and our step-by-step guide to getting recommended covers the rest. For the reference version of how AI search works, see the wiki guide.