When ChatGPT chose which local businesses to show in our study nothing came close to the pull of an address inside the asked city.

Every owner who hears that ChatGPT recommends local businesses asks the same follow-up: what is it looking for? Our August 2026 study gives a measured answer for one week and one surface. We asked "who is the best {service} in {city}" across 216 US markets, 12 service types in 18 cities, recorded the ChatGPT web app's own responses, and compared the businesses it showed against the candidates it observed but left out.

One caution up front, because this space is full of overclaiming. What follows is descriptive. It says which traits were more likely to be shown in this study, with confidence intervals attached. It does not say ChatGPT weighs these things, only that the businesses it displayed looked like this. The full model and its limits are on the study page at how ChatGPT picks a local business.

The study compared shown businesses against their competition

Across both waves, ChatGPT's answers exposed 6,024 candidate businesses, of which 1,946 were actually shown as recommendation cards, across 428 answer strata. Because we could see candidates that did not make the cut, we could ask the interesting question: within the same answer, what separated the businesses that got a card from the ones that did not?

We fit a conditional logit, which compares businesses only against the other candidates in their own answer, and expressed the results as odds ratios. An odds ratio of 2 means a trait was associated with double the odds of being shown, holding the rest of the model constant.

The in-city address towered over every other trait

Nothing else was close. An address inside the asked city carried an odds ratio of 14.4 (CI 10.3 to 21.5).

The raw shares make it vivid: 92.0% of shown businesses had an in-city address (1,790 of 1,946), versus 73.9% of observed candidates that were not shown (3,020 of 4,086). Out on the answer surface, only 83 of 975 displayed cards, 8.5%, came from outside the asked city, though 52 of 215 panels, 24.2%, contained at least one such business.

For multi-location operators and service-area businesses, this is the finding to sit with. In this data, ChatGPT behaved as if the city in the question is a hard filter it strongly prefers to satisfy with businesses literally addressed there. Serving a city and being addressed in it read as different things.

Reviews and ratings raised the odds of being shown

After location, the familiar reputation numbers did the work:

• Review count: each doubling was associated with 1.81 times the odds of being shown (CI 1.71 to 1.94). Going from 50 to 400 reviews is three doublings.
• Rating: each additional 0.1 star was associated with 1.36 times the odds (CI 1.26 to 1.49). Steep, because the competition is compressed near the top.
• Yelp-sourced rating on the card: odds ratio 3.31 (CI 2.35 to 4.46), which ties into the Yelp logo tell from earlier in this series.
• A service word in the business name: odds ratio 1.23 (CI 1.02 to 1.50), a modest edge for names that say what the business does.

And one welcome null: GBP category tags showed an odds ratio of 1.02 per tag (CI 0.99 to 1.05), statistically indistinguishable from nothing. Stuffing extra categories onto a profile bought no measurable lift in this data. That said, relevance itself was visible: 64.4% of displayed cards carried the asked service among their GBP categories (628 of 975).

The entry bar sits near a 4.8 rating everywhere

Odds ratios describe the race. The entry bar describes the ticket price. Looking at the weakest business on each displayed panel, the median panel-minimum rating was 4.8 in every market tier we measured. No tier relaxed it.

Review counts scaled with market size: the median panel-minimum was 72 reviews in major metros, 95 in mid-sized metros, 62 in suburbs, and 50 in small cities. Across the 18 study cities, city population and the review bar moved together (Spearman rho 0.53, exploratory). Small markets are the friendlier on-ramp, and 4.9-rated businesses with a few hundred reviews are the standard company you keep on these panels.

Two hygiene findings round it out: across the study, zero closed businesses were displayed (0 of 1,959 closed candidates), and chains barely registered, with only 5 of 969 displayed businesses appearing in more than one study city (0.5%). These panels are overwhelmingly open, local, independent operations.

What a business can do with descriptive numbers

Held next to the churn finding, a coherent picture forms. Which businesses appear varies ask to ask, but the profile of what gets shown was stable: addressed in the city, rated near the top, reviewed well past the local bar. You cannot force any single appearance. You can make your business match the profile that kept getting drawn.

Practically, that means leading your presence with the city you are actually addressed in, building review volume past your market's bar with a rating that can stand next to 4.8s, and skipping the tag-stuffing theater the data says bought nothing.

The questions owners ask most about what ChatGPT looked for in local businesses, answered from the study data.

Ready to measure your business against the shown profile?

The study hands you a checklist you can score yourself against today: in-city address for the markets you care about, rating that holds up beside a 4.8 bar, review count past your market's median minimum, a clean profile on both rating sources. Spearleaf runs that exact comparison, business by business, inside our AI search work. If you want to see where you clear the bar and where you are giving away odds, reach out through our contact page and we will score it with you.

This article was written by Joshua Albanese, founder of Spearleaf, a local SEO and marketing agency based in Fort Myers, Florida. Joshua has built six businesses from zero using organic search and content, and he leads Spearleaf's SEO, AI search, and local strategy for service businesses across the U.S.