Why AI recommends one of your locations and ignores the rest

A regional HVAC company with twelve branches asks an assistant “who does emergency AC repair near me” in each of its twelve towns. In four towns, its own branch gets named. In three, a single-location competitor gets named instead, one with a fraction of the reviews and none of the brand’s national advertising. In the rest, nothing local gets named at all, just generic advice about calling a licensed technician. Same brand, same training, same call center behind every branch. Twelve different answers.

This is not a fluke of one bad month. It is the normal, measured behavior of AI answers for any business with more than one location, and it means a single “AI visibility score” for a multi-location brand is close to meaningless. Reporting one number when the real answer varies by town is the same mistake as reporting one national ranking when the real business runs on local demand.

Independents start ahead of you, structurally

The clearest data on this comes from SOCi’s 2026 Local Visibility Index, the first large-scale study built specifically to measure how often multi-location brands get named in AI answers rather than just where they rank. Across roughly 350,000 locations and 2,751 multi-location brands, ChatGPT recommended only 1.2% of the individual locations analyzed, Perplexity recommended 7.4%, and Gemini recommended 11%, against 35.9% visibility for the same brands in Google’s local 3-pack - the study’s authors put AI assistants at up to 30 times more selective than Google’s local results.

That gap is not a sign the brand is doing worse work than the independent salon or repair shop next door. It is what happens when an engine has to name two or three businesses instead of showing a ranked list of ten. A tighter shortlist rewards the business with the fewest reasons to hesitate, and independents tend to have fewer moving parts to get wrong.

The reason is entity-level, not store-level

The part that matters most for a multi-location brand is why the shortfall happens, and it is not evenly spread across branches for reasons a store manager controls. SOCi’s own writeup of the same index found that brand locations appeared in AI recommendations only 6.5% of the time overall, against 36% in Google’s 3-Pack, and traced the cause to how these systems build a single opinion of a brand from every location, platform and review at once, then apply that opinion everywhere. A handful of neglected profiles, stale hours, or thin review pages can drag down a location that has done everything right, because the engine is not really evaluating that location in isolation. It is evaluating the brand, using that location as one data point among many.

This is the opposite of how classic local search worked. A ranking algorithm could still surface your best-optimized page even if your worst one was a mess. An assistant deciding who to name is making a trust judgment about the whole entity, which means your weakest location’s data quality is doing damage well beyond that one town.

The answer also is not stable, so testing once tells you nothing

Even setting aside the brand-versus-independent gap, the answer for a single query moves more than most people assume. Steady Demand’s citation analysis, published in August 2026, ran 1,487 local search queries across 50 U.S. metro areas and 10 service categories, pulling 14,472 citations from Gemini and finding that identical queries returned overlapping sources only about 40% of the time, compared with roughly 90% for Google’s local pack; the same top business was named again in only about 7% of repeated searches, and Gemini and ChatGPT agreed on the top business just 4.2% of the time.

Put those two findings together and the practical conclusion is unavoidable: if you check one town, on one engine, once, you have learned almost nothing about the other eleven towns, or about what the same engine will say next week. A flagship location that gets named consistently tells you that flagship is healthy. It tells you nothing about whether branch nine is showing up at all.

Data quality decides who gets trusted enough to name

Part of why Gemini recommends more often than ChatGPT or Perplexity traces back to something boring and fixable: business profile data was measured at only 68% accurate on ChatGPT and Perplexity, compared with 100% on Gemini, largely because Gemini leans more heavily on Google Business Profile data that brands already keep current for local search. An assistant that cannot confirm your hours, address or category with confidence has an easy, safe answer available: name someone else, or name no one. Our guide on getting a local business recommended by AI assistants covers the specific fields worth locking down first, and the same checklist applies per location, not once at the brand level.

Fix it location by location, not brand-wide

The fix that does not work is a single push at headquarters: update the main site, run one PR campaign, call it done. Every location needs its own genuinely distinct evidence, because a templated page with the town name swapped reads to an engine the same way it reads to a person, as filler. That means a real address and service area stated in the page text, reviews that mention what was actually done and where, and local specifics an assistant can quote directly rather than a paragraph of brand copy repeated twelve times. What actually decides which sources an engine is willing to cite applies at the location level exactly as it does at the brand level, and it is worth auditing each location’s page against it separately rather than assuming the flagship’s strength covers the rest.

Track it like it’s twelve businesses, because it is

Given how much the answer varies by town and drifts over time, a monthly brand-level percentage is close to useless for deciding where to spend effort next. The honest version of this report tracks the actual local questions per town, on the engines that matter for that market, and treats each location’s result as its own line rather than folding it into a company-wide average. That is also the only way to catch a location quietly losing ground to the competitor named in its place, the same pattern covered in tracking competitors in AI answers. And because the underlying answer set shifts on its own even without any competitor doing anything differently, per our look at what actually changes when a model updates, a location that dropped out of the answer needs a re-check before anyone concludes the branch did something wrong.

See what each of your locations actually says

Pick three towns you operate in and ask an assistant who it would recommend in each. If the answer is not your business in all three, you have just seen, in miniature, what the research above found at scale: multi-location brands do not lose visibility evenly, and they do not lose it for the reasons a regional manager would guess. Visibility AI’s monitoring tracks each location’s questions separately across engines, and the free Visibility Check is the fastest way to see where a specific town stands today.