How AI decides which local business to recommend

A customer used to type “dentist near me,” see ten blue links, and choose. Today a growing share of them ask ChatGPT, Perplexity, or Google’s AI answer the same question and get back a sentence with one or two names in it. The list is gone. The recommendation replaced it.

At Local AI Registry we measure this for a living. We ask AI assistants the questions real customers ask, record which businesses they name, and score how findable and recommendable each one is on a 0-to-100 scale we call the AI Score. The finding that should worry every owner: across the local businesses we have measured so far, more than three in four are named by no AI assistant at all. Fewer than one in four are recommended by even one.

And the businesses that do get named are rarely the ones with the nicest website. They are the ones the rest of the web agrees on. That single recommendation is worth more than any page-two ranking ever was: the business that wins it takes the booking, and the ones that lose it are never seen. So the question every owner should be asking has changed from “where do I rank?” to “does AI recommend me, and why not?”

An AI assistant does not answer a local question from memory. When you ask it for a recommendation, it retrieves: it goes out to the live web and its search index, pulls a set of candidate businesses and the sources that describe them, and then generates an answer grounded in what those sources say. This process is called retrieval-augmented generation, and understanding it is the whole game.

This is documented, not guessed. OpenAI describes how ChatGPT rewrites your question into targeted searches before it answers, and Google describes a query fan-out that issues multiple related searches to build an AI Overview. Different labels, the same shape: retrieve first, then answer from what was retrieved. Two parts of that process decide the outcome, and they are the two signals we weight most heavily in the AI Score.

Retrieval grounding

Retrieval grounding is whether the assistant can find and verify your business for the specific question that leads to a booking. A med spa can rank for its own brand name and still be invisible for “best med spa in Costa Mesa,” which is the query that actually converts. Grounding measures presence on the money question, not the vanity one. If the assistant cannot retrieve you for the query that matters, nothing else you do can save you.

Off-property attestation

Off-property attestation is what third parties say about you: reviews, directories, “best of” lists, local press, and mentions across the web. Assistants trust corroboration. A claim you make on your own homepage is marketing; the same claim made by a dozen independent sources is evidence. Attestation is the signal owners most often ignore and the one that most often decides the recommendation.

We do not keep the method a secret. The AI Score is built from eight signals, and every weight is published. In rough order of how much they move an AI recommendation:

• Retrieval grounding: can AI find and cite you for the queries that drive bookings.
• Off-property attestation: what the wider web says about you, independently.
• AI visibility: whether specific assistants already name you today.
• Reviews: volume, recency, rating, and whether you respond.
• Google Business Profile: completeness and accuracy of your listing.
• Authority: the quality of the sites that link to and reference you.
• Speed and mobile: whether your site is technically eligible to be cited.
• Content and schema: whether your pages are machine-readable and clear.

Here is the counterintuitive part. You can have a fast, beautiful, keyword-perfect website and still be the business AI never mentions. An assistant weighs what independent sources say about you far more than what you say about yourself, because independent sources are harder to fake. Your homepage is a claim. The web’s consensus is the evidence the assistant grounds its answer in.

This is why two businesses on the same street, with similar websites, get opposite results. The one that shows up in review platforms, gets named in local roundups, and carries consistent details across directories is the one the assistant can corroborate, so it is the one that gets recommended.

AI recommends the business the web agrees on, not the business with the nicest homepage.

The mechanism is shared, but each assistant sources local recommendations a little differently, so a complete strategy accounts for all of them.

• ChatGPT browses the live web and leans on well-structured, well-cited sources. Clear content and strong attestation help most here.
• Perplexity is citation-first and shows its sources in the answer, so off-property attestation is close to everything.
• Google AI Overviews draw heavily on the local pack, the Google Business Profile, and reviews, so your Google presence carries extra weight.
• Gemini leans on Google’s knowledge graph and your Business Profile, which rewards consistent, complete listings.

These are not folk theories. OpenAI documents that ChatGPT shows inline citations and a sources panel in its answers, and Google’s own guidance tells local businesses to keep their Business Profile and Merchant Center current to be eligible for AI features. The differences show up in our data too: in our checks so far, Perplexity named local businesses most often and Gemini least.

The order of operations is clear once you know the mechanism. First, find out what AI already says about you. Then fix the two signals that decide the recommendation before you touch the rest.

• See where you stand: get your free AI Score and read the specific signals holding you back.
• Understand the method: our scoring methodology documents all eight signals and their weights.
• See who AI recommends near you: browse the registry leaderboards for your category and city.
• Know it is honest: we never sell rank, so a high score means the web genuinely agrees on you.

AI local search is early, and it moves quickly. The businesses that understand the mechanism now, and build real attestation, will be the names assistants return for years. The ones waiting for it to matter will find the recommendation already went to someone else.