Local businesses
Does AI recommend your business?
45% of consumers now ask AI for local recommendations. SOCi measured ChatGPT recommending just 1.2% of the 350,000 US local businesses it tracked. We test the questions your customers ask, show you whether you make the answer, and hand you a fix list.
The recommendation set is short, and mostly chains
When someone asks an AI engine for a local recommendation, the engine draws on two layers: the live local layer (Google local pack data, review sites) and a trained layer (what the model already believes about a category in a city).
Lightspeed's study with Vaer found 90 to 94% of AI answers to local questions run in training-data mode rather than live web results. That explains the pattern SOCi's measurement shows: ChatGPT recommending just 1.2% of 350,000 US local businesses, roughly 12 of every 1,000.
The trained layer favors chains, because chains carry the largest text footprint. In Lightspeed's data, Google AI Overviews named no local store in 68% of responses, and the chain share of answers only fell from 44% to 9% when users explicitly asked for independent options.
Sources: SOCi data (vendor research) and Lightspeed x Vaer study (vendor research; Lightspeed sells to local businesses).
BrightLocal reports 45% of consumers now use AI for local business recommendations, up from 6% in earlier tracking. The audience asking AI for local recommendations is real and growing.
Recommended by ChatGPT
of 350,000 tracked US local businesses
AI answers drawing on Google's local top three
SOCi overlap measurement
Traditional + AI answer overlap
what both layers agree on
Restaurants absent from AI answers
the category studied
All figures: SOCi, tracking 350,000 US local businesses across AI answers. Vendor research, labeled as such.
What the research shows
The published numbers on local AI recommendations, with sample and publisher attached to each figure, because vendor research and consumer surveys travel different distances.
45%
of consumers now use AI for local business recommendations, up from 6% in earlier tracking. Consumer survey data, not vendor projection.
68%
of Google AI Overviews for local-intent queries named no local store at all. Lightspeed x Vaer, 460,000 responses.
44% → 9%
Chain share of AI answers when users asked generally (44%) versus explicitly for independent options (9%). The question shape changes the recommendation set.
1.2%
ChatGPT recommended just 1.2% of the 350,000 US local businesses SOCi tracked. Roughly 12 of every 1,000.
What customers ask AI
Example questions, illustrative"Best dentist near me that's open Saturdays"
"Where should we hold a 30-person offsite in Austin?"
"Family doctor accepting new patients, not a hospital system"
"Best sushi on the Portland waterfront"
"Affordable dog groomer that takes one-time visits, no package"
Constructed examples based on common local questions, not captured from real sessions. The Written Diagnosis replaces them with the questions your actual customers ask.
How the answer looks when you're skipped
"Here are three well-known names in that area: [Chain A], [Chain B], and [Chain C]. I couldn't find enough reliable information about smaller independent businesses there."
Pattern we see repeatedly: independent businesses aren't rejected, they're never assembled into the list, because the trained layer holds only the names with the largest text footprint.
Six checks you can run today
No tools needed beyond the engines themselves. Write down what you see each time; the pattern across answers is the finding.
Open a fresh ChatGPT session and ask as a customer would: "Best [category] in [your city]?" Note whether your business is named, and which names come up instead.
Ask "Where should I eat in [city]?" and the same question with "independent, non-chain" added. Compare the two answers. That gap is the trained-layer problem in one exchange.
Check your Google Business Profile is complete: categories, hours, services, review responses. SOCi found 35.9% of AI answers drawing on the Google local top three, so the pack feeds the AI layer.
Search your category plus your city and see which review sites and local guides appear. Those are the third-party sources the trained layer reads; your absence there compounds.
Check whether your homepage states your city, your category, and your hours in plain text the engines can quote. If it's only in images or embeds, you're invisible to that layer.
Repeat the same question a week later. Single runs mislead; the pattern across engines and time is the real signal.
From free check to written diagnosis
The free check runs against your homepage and reports seven technical aspects that decide whether AI engines can read your site at all. It is not local-specific, and it says so on the result.
The Written Diagnosis ($490) answers the local question for your category and city: 30 buyer-style questions across ChatGPT, Perplexity, and Gemini, run twice over two days, written up as a report of who gets recommended in your area, where you stand, and what to fix first. The Fix Sprint ($1,250) executes that fix list for you. Monthly monitoring tracks how the answers shift as engines update.
Keep reading
Questions local businesses ask
Do customers really ask AI for local recommendations?
BrightLocal's survey work reports 45% of consumers now use AI for local business recommendations, up from 6% in earlier tracking. The habit is new, growing fast, and already at 45%.
Why does AI recommend so few local businesses?
Because the recommendation layer is mostly trained, not live. SOCi measured ChatGPT recommending just 1.2% of the 350,000 US local businesses it tracked. Lightspeed's study with Vaer found 90 to 94% of AI answers in training-data mode for local questions, and Google AI Overviews named no local store in 68% of responses. The engines default to a short list that was set somewhere else.
My Google profile is in good shape. Isn't that enough?
It helps, and it is not the whole layer. SOCi found 35.9% of AI answers drawing on the Google local top three, so local-pack health feeds the AI layer. But with 1.2% recommended overall, the AI layer selects from a much shorter list than the map pack shows. The two layers reward overlapping but different things.
Can a small independent business get into AI answers?
The question shape matters. Lightspeed's study found that when users explicitly asked for independent (non-chain) options, the chain share of AI answers dropped from 44% to 9%. Asking differently changes what comes back, which means what the engines can learn about you matters. Vendor-published research, labeled as such.
What does the free check cover?
The free check examines seven technical aspects of your homepage instantly and shows the result of each. It is not local-specific: it tells you whether AI engines can read your site at all, a precondition for being recommended. No signup required.
See whether the engines can read your business
The free check runs instantly against your homepage. No signup, no cost.
Run the free check