How to get recommended by AI for your local business

We asked AI assistants, with live web search, the questions people actually type when looking for a local business — across 61 city and category pairs. This is what the answers are built from, and what that means for you.

What actually changed

A search engine gives you ten blue links and lets you choose. An AI assistant gives you three names and a reason for each. There is no second page, no "see more results", and nobody scrolls looking for a fourth option. You are either one of the three or you are not in the conversation.

This is not a hypothetical shift. BrightLocal's 2026 consumer survey found that 45% of people had used ChatGPT, Gemini, Perplexity or a similar assistant to find a local business in the past year, up from 6% the year before. Meanwhile SOCi's 2026 index reported that ChatGPT recommended only about 1.2% of the locations it analysed. Both of those numbers come from companies that sell in this space, so treat them as directional — but the direction is not in doubt.

Where the answers come from

This is the part most advice gets wrong. An assistant answering "best dentist in Austin" is not consulting a ranking. It runs a search, reads a handful of pages, and summarises them. So the question is not "how do I rank" — it is which pages does it read.

We counted the sources cited across our measurements. The pattern repeats city after city, category after category:

Source type	What we saw
Reddit	The single most cited domain, by a wide margin. In one Austin dentist measurement it was cited 92 times across 8 answers.
Local blogs and city guides	Small, specific sites — a food blog, a "things to do in" site, a neighbourhood guide. Cited far more than their traffic suggests.
Category directories	Zocdoc, Opencare, Yelp, TripAdvisor and niche-specific lists, depending on the category.
The business's own site	Cited, but usually only after the business has already been named somewhere else first.

That last line is the uncomfortable one. Your own website rarely gets you into the answer. It confirms you once something else has put you there.

What this means in practice

Be where the model reads. Find the sources cited for your category in your city — not in general, in your city — and get a legitimate presence on them. That list is different for a dentist in Austin and a roofer in Denver.
Earn real mentions in community threads. Reddit's weight is hard to overstate. That does not mean astroturfing: fake recommendations get detected, get removed, and poison the well. It means being the kind of business people name when someone asks.
Make your details consistent everywhere. Same business name, same address, same phone, on every listing. Models discard contradictory information rather than guess.
Keep reviews recent and answered. Recency is cited constantly in the summaries we read; a wall of five-year-old reviews reads as a closed business.
Say plainly what you do and where. A page that states the service and the neighbourhoods, in plain sentences, is quotable. A page built entirely of images and slogans is not.
Measure it monthly. These answers move. A position you held in October can be gone in November, and nothing tells you unless you look.

What does not work

There is no submission form for AI assistants, no "AI schema" that forces an inclusion, and nobody can sell you a guaranteed position. If someone offers that, they are selling you something they cannot deliver. What exists is measurement, a source list, and the ordinary work of being present on those sources.

See where you stand, free

We ask an assistant the questions your customers actually type, with live web search on, and tell you whether you get named, who gets named instead, and the exact sources behind those answers.

Check my business

The public measurements

We publish what we measure. Browse the index to see who AI recommends by city and category — dentists in Austin, plumbers in Denver, gyms in Tampa, dentistas en Madrid, and more.