Foursquare Powers ChatGPT's Map — But Not Its Text Recommendations

OpenAI confirmed that Foursquare provides the data behind ChatGPT's map cards -- those little location pins that appear when you ask about local businesses. So naturally, everyone assumed: get on Foursquare, get recommended by ChatGPT.

We tested it. Foursquare presence has a -17% correlation with appearing in ChatGPT's text recommendations. Negative seventeen percent. Businesses on Foursquare are less likely to show up in ChatGPT's written answer than businesses that aren't.

That sounds contradictory until you realize ChatGPT has two completely separate recommendation systems. They don't talk to each other.

System 1: Text recommendations. When ChatGPT writes "I'd recommend these plumbers in Manchester..." and lists businesses with descriptions and links, that comes from Bing web search. ChatGPT queries Bing, reads the results, and synthesizes an answer. Foursquare is nowhere in this chain.

System 2: Map cards. The visual map pins that sometimes appear in the ChatGPT interface come from Foursquare's places database. This is a separate data feed rendered by the UI, not something the language model generates.

These two systems frequently recommend different businesses for the same query. The text might name five plumbers while the map pins show three completely different ones. A user sees both and assumes they're from the same source. They're not.

The -17% correlation probably exists because businesses on Foursquare tend to be the ones with strong Google Business Profiles and local pack presence. These businesses built their online presence around platforms that feed Google's local results. But ChatGPT's text system doesn't access Google's local pack at all. It goes through Bing.

We confirmed this by testing Foursquare's API directly. We could replicate the map card results by querying Foursquare ourselves -- the same businesses appeared. But those businesses had minimal overlap with what the text recommendation system surfaced.

This is a genuinely confusing setup. One product, two recommendation engines, different results. And the one that most users read -- the text answer -- is the one that Foursquare doesn't influence.

Does this mean you should ignore Foursquare? Not necessarily. The map cards are still visible to users. Having a pin on the map isn't worthless. But if you're registering on Foursquare specifically to improve your chances of being named in ChatGPT's text answer, our data says that's the wrong lever.

The things that correlate with text recommendations are: having a website (+44 percentage points), social profiles linked on your site (Facebook +42pp), and being listed on directories that Bing indexes heavily -- Yelp, Trustpilot, Checkatrade.

Foursquare is useful for Foursquare-powered features. For the part of ChatGPT that actually writes out recommendations, it doesn't register.