1.2 Percent: The New Odds of Getting Recommended by AI Search
When The Data Goes Dark
A nonprofit that tracks meteors for astronomers got knocked offline by a cyberattack this year. Not a huge story. It barely made the news outside of a few science blogs. But the group ran on a small number of volunteer-fed sensors, and once the network went down, decades of sky data just stopped updating. No backup source picked up the slack because there wasn't one.
Local business data works the same way, and most business owners have no idea.
You assume your information lives in a hundred places. Google Business Profile, your website, review sites, directories you filled out once in 2019 and forgot about. Feels redundant. Feels safe.
It is not redundant. AI platforms pull from a small handful of sources, and if your data is thin or wrong in those specific places, you do not exist to the model asking the question on your behalf.
Foursquare supplies roughly 70 percent of the local business data ChatGPT uses, according to reporting from April 2026. Yelp accounts for 62 percent of Perplexity's local citations, based on a July 2026 analysis of 28 million responses. Two platforms, most of the answer.
That is not a diversified pipeline. That is a single point of failure with a friendly interface on top of it.
The Odds Are Not In Your Favor
ChatGPT recommends about 1.2 percent of local business locations when it answers a question. Not 1.2 percent of the ones that ask nicely. Not 1.2 percent of the ones with bad websites. 1.2 percent of all locations, period.
Perplexity does better, recommending 7.4 percent, per the same index. Still a rounding error compared to what most owners think their odds are.
Compare that to Google Maps, which still surfaces a business in 66 percent of local searches. BrightLocal ran the numbers on this in September 2026, testing 200,000 localized AI queries, and found AI platforms landing somewhere between 32 and 38 percent. Maps and AI recommendation are not measuring the same thing, even though they look like cousins.
Ranking well on Maps got you into the conversation for twenty years. Being one of the businesses AI decides to name out loud is a different contest, with a much smaller number of seats. You can hold the top Maps ranking in your zip code and still be one of the 98.8 percent ChatGPT never mentions.
That gap is the whole story right now. Owners keep optimizing for the game with 66 percent odds while the game that matters is closer to a lottery.
Who Actually Feeds The Machine
Foursquare and Yelp did not become local data authorities by accident, and they did not do it recently. Foursquare has been quietly powering location data behind the scenes for over a decade, serving check-in apps and mapping tools long before anyone cared about AI answers. Yelp built its review corpus one restaurant complaint and one five-star rave at a time, for years, before a single chatbot ever cited it. The models did not build new infrastructure. They plugged into what already existed and was already deep.
That matters for what you actually do this week. If your business listing on Foursquare has an old address, a disconnected phone number, or a category that does not match what you actually sell, ChatGPT is working from bad source material and does not know it. Same with Yelp. A thin review profile with three vague comments gives Perplexity almost nothing to extract when it is deciding who to name in an answer.
Google Business Profile still matters, obviously. But treating it as the only listing worth maintaining is exactly the assumption that keeps businesses out of the 1.2 percent. NAP consistency, name, address, phone, has to hold across Foursquare and Yelp specifically, not just Google. Detailed reviews that mention specifics, services, neighborhoods, what a visit actually involved, give these platforms something concrete to pull from. Vague five-star reviews with no text are close to invisible.
What Changed This Week
On September 17, Google started converting local knowledge panels into AI Overviews. Not testing it in a corner of Maps. Rolling it into the surfaces businesses have relied on for years to get found.
This is Google telling you where the puck is going. Knowledge panels used to be a passive lookup, hours, phone number, a map pin. AI Overviews generate a synthesized answer, which means Google is now doing to its own local results what ChatGPT and Perplexity already do to the open web: reading, extracting, and deciding what to say on your behalf instead of just showing what is there.
BrightLocal's September 16 study of 200,000 localized AI searches found website citation mattered more than expected in these results. Not just profile completeness. The actual content on your site, structured in a way a model can lift and use.
The practical response has not changed, but the deadline just moved closer. Schema markup on your website tells these systems what they are looking at. Extractable content, actual service details, actual pricing context, actual specifics, gives them something to cite instead of guess at. And your profiles need to match across every source feeding these models, not just the one Google happens to be rebuilding this week.