Every AEO guide published in the last year repeats the same claim: 60 to 70 percent of ChatGPT's local business results come from Foursquare. Contractors are told to claim their Foursquare listing, complete it, and keep it current. That advice sounds reasonable. The problem is that the underlying claim does not hold up when tested at scale.

A study by SteadyDemand in August 2026 ran 2,880 prompts designed to give Foursquare every reasonable advantage. The result: 0% Foursquare citation share on ChatGPT's primary surface. Not 60%. Not 70%. Zero.

Where the 60-70% number came from

The original finding was real but narrow. A Spanish-language test used 50 prompts across five Spanish cities and inspected ChatGPT's raw JSON responses. In that specific sample, 60 to 70 percent of the businesses ChatGPT placed first came from Foursquare data. That was May 2025, first-ranked results only, five cities in Spain.

Seven of eight downstream articles that cited the finding dropped every one of those caveats. "60 to 70 percent of ChatGPT's local results come from Foursquare" became the headline. The scope, the market, and the ranking position all disappeared. A finding about first-ranked Spanish restaurant recommendations became a universal claim about all local search everywhere.

The same OpenAI-Foursquare partnership that generated the original finding is real. Foursquare's Places API contains more than 100 million points of interest and does power ChatGPT's search. A partnership existing and a partnership driving the majority of visible results are different things, and the data says the second claim is false.

What ChatGPT actually reads

ChatGPT assembles a local answer from two layers, built by different suppliers. Most businesses manage only one of them.

The first layer is the sentence. This is the text that names the business and describes what it does. It comes from editorial articles, directory roundups, local blog posts, and forum threads. ChatGPT reads these to decide which businesses to mention at all. In the Pluspoint study that ran 110 dining queries across 11 US cities, city and national editorial supplied 48.9 percent of all cited domains. Community discussion supplied roughly one in ten more. Not one answer out of 210 read a Google, Bing, Apple, or Yelp page to decide its picks.

The second layer is the card. This carries the address, photos, and booking link that appear alongside the sentence. In SteadyDemand's audit of 56 structured business cards, 45 carried a Yelp URL tagged as an OpenAI data feed. Google, Apple, and Bing appeared on none of them. Yelp occupied the card slot in 95.83 percent of observed cases.

That split explains a failure pattern that contractors keep running into. Rewriting a website moves neither layer. Foursquare listing work improves the card but leaves the sentence alone. The sentence moves when other people write about the business, and the sentence decides whether the customer ever reaches the card.

The overlap test

SteadyDemand compared ChatGPT's named businesses against ranked sets from both Google and Foursquare for the same markets and categories. If Foursquare were the stronger hidden influence, ChatGPT's picks should resemble Foursquare's more. They did not.

Mean overlap with Google was 7.5 percent. Mean overlap with Foursquare was 5.7 percent. Google resemblance was significantly higher across every market tier tested. Dense urban markets, mid-size cities, and small towns all showed the same pattern. Foursquare did not win anywhere.

What contractors should do instead

The 2,880-prompt study is a single snapshot from August 2026, four weeks after a competing licensing announcement. Data partnerships shift, and a future OpenAI deal could change the card-layer supplier. But the contractors spending time on Foursquare today are spending it on a surface that showed zero citations in 2,880 tests.

Here is what actually moved the needle in the same research.

Keep Google Business Profile current. Google showed higher resemblance to ChatGPT's picks than any other source. GBP is the single most important entity signal for local AI recommendations. Business category, service area, hours, photos, and the services list all feed the data that ChatGPT reads.

Build reviews across Yelp and Google, not just Google. Yelp occupied the card slot in 95.83 percent of structured responses. A trades business with 50 Google reviews and zero Yelp reviews is invisible on the card layer. Send every third or fourth review request to Yelp instead of Google.

Earn mentions in local editorial and community forums. The sentence layer comes from articles and discussions written by other people. A local roundup of "best HVAC contractors in [county]" or a Reddit thread where a homeowner describes a good experience gives ChatGPT something to quote. Businesses that appear in that kind of content get named. Businesses that only exist in directory listings do not.

Make the description matchable. ChatGPT matches the query against the words in reviews and articles, not against star ratings. A plumber whose reviews say "fixed my burst pipe in 20 minutes on a Sunday" gets matched when someone asks ChatGPT for an emergency plumber. A plumber whose reviews all say "great service" does not. Detail in the review text matters more than the rating.

Why the myth persists

The Foursquare claim circulated because it gave AEO consultants something concrete to sell. "Claim your Foursquare listing" is a clear action item. "Earn mentions in local editorial" is harder to package. But the data is unambiguous: in 2,880 tests, Foursquare contributed zero citations to ChatGPT's primary surface. Yelp contributed 95.83 percent of the structured cards. Google showed higher resemblance to ChatGPT's picks than any other source.

Contractors who want to show up in AI answers should spend their time on the surfaces that actually feed the model. That means Google Business Profile, Yelp reviews, and local editorial mentions. Not Foursquare.

FAQ: ChatGPT Local Business Recommendations

Does ChatGPT use Foursquare data to recommend local contractors?

A 2,880-prompt study by SteadyDemand in August 2026 found 0% Foursquare citation share on ChatGPT's primary surface. Yelp appeared on 95.83% of structured business cards instead. Foursquare has a real OpenAI partnership, but the widely cited 60-70% claim traced back to a 50-prompt test across five Spanish cities that only measured first-ranked results.

What data does ChatGPT actually read when recommending local businesses?

ChatGPT reads two layers. The sentence, which names the business, comes from editorial articles, directories, roundups, and forum threads. The card, which carries the address and booking link, comes from a licensed business-data feed, with Yelp occupying that slot in 95.83% of observed cases.

Should contractors update their Foursquare listing for AI visibility?

Foursquare listings do not hurt, but the August 2026 data shows 0% citation share on ChatGPT's primary surface. Contractors get better results from keeping Google Business Profile current, building reviews across Yelp and Google, and earning mentions in local articles and community forums.

What has a higher overlap with ChatGPT local picks, Google or Foursquare?

Google. The SteadyDemand study measured mean overlap between ChatGPT's named businesses and Google at 7.5%, versus 5.7% for Foursquare. Google resemblance was significantly higher across every market tier tested.

How can a trades business check what ChatGPT says about them?

Run 10 to 15 relevant local queries without including the business name, twice each, and document whether the business appears in the answer. Check what the response actually says about the business. A single ChatGPT screenshot proves nothing. Run the same queries monthly to track movement.

Run your free AI visibility audit to see what ChatGPT, Gemini, and Perplexity actually say about your trades business.