Short answer: in an August 2026 test, ChatGPT’s local business cards were backed by Yelp, not Foursquare. Steady Demand ran 2,880 prompts across 12 metro areas and 12 business types. It found Yelp attached to the structured business cards in 95.83% of runs, while Foursquare appeared in 0.00% of citations on ChatGPT’s main surface.
If you run a restaurant, a plumbing company, a dental office, a salon or a store in Fort Wayne, Auburn or anywhere in Northeast Indiana, that matters. For more than a year, local SEO articles have told owners that most of ChatGPT’s local results come from Foursquare listings. Some owners have spent real time on that advice. This test says the advice was aimed at the wrong listing site, at least for now.
This post covers what the test measured, where the Foursquare claim came from, what Yelp’s role means, what the restaurant-specific data says, and a short fix list. We also say plainly where the evidence is thin. It is one snapshot, taken by a third party, in a fast-moving market.
Key Takeaways
- Steady Demand’s August 2026 test (2,880 prompts, 4,607 successful runs) found Yelp in the structured business-card data of 95.83% of ChatGPT local answers and Foursquare in 0.00% on the main surface.
- The popular “60 to 70% from Foursquare” claim traces to a May 2025 study of 50 prompts in five Spanish cities, first-ranked result only. The limits fell off as the number was repeated.
- Yelp’s share ranged from 93.75% for hair salons to 97.92% for plumbers, so it was not limited to restaurants or “Yelp-friendly” categories.
- For restaurants, TripAdvisor (28.90%), OpenTable (16.88%) and Resy (8.96%) also showed up in business-card data in that test.
- Diners still check before they act. In a Menufy survey of 1,000 U.S. consumers, only 2% would order from an unfamiliar restaurant an AI recommended without checking another source first.
- This is a snapshot four weeks after a reported, non-exclusive Yelp licensing deal. It could change, so recheck quarterly.
- Fix list: claim and complete your Yelp page, match it to your Google Business Profile and website, keep reviews current, and stop spending time on Foursquare for AI reasons.
What Did the August 2026 ChatGPT Test Actually Find?
Steady Demand published the test. The design is worth a close look, because it was built to give Foursquare every advantage. Seven of the 12 business types came from Foursquare’s natural territory: restaurants, bars and nightlife, coffee shops, hotels, museums and attractions, live-music venues, and bakeries. The 12 metro areas included dense cities such as New York, San Francisco, Chicago and Los Angeles. Plumbers and auto repair stayed in as unfavorable controls.
ChatGPT’s main surface returned usable data for 2,879 of the 2,880 prompts. A second set of 1,728 prompts was collected separately on a different API surface. Together that is 4,607 successful runs, all collected within one 72-hour period in August 2026.
The researchers measured three layers: domains cited in visible links, domains retrieved but not shown, and URLs attached to structured business cards. On the main surface, Foursquare-controlled domains appeared in 0.00% of all three layers across all 2,879 runs, with a 95% confidence interval of 0.00% to 0.13%. On the second surface, Foursquare showed up once in 1,728 runs. That one citation was a place-redirect page for a gym in Los Angeles, a rate of 0.06%.
Our reading, which is judgment and not data: a test designed to favor Foursquare that still found almost none is hard to wave away. Steady Demand is careful on one point, and so should we be. Citation share is not the same as data influence. Zero visible links do not prove Foursquare data never touches an answer. The researchers tried to check that indirectly, and we cover it in the next sections.
One more point for accuracy. OpenAI and Foursquare do have a real partnership, and Foursquare says its Places data covers more than 100 million points of interest. Nothing in this study disputes that. It tests a narrower claim, that 60 to 70% of ChatGPT’s local results come from Foursquare city-guide listings.
Where Did the “60 to 70% Foursquare” Claim Come From?
Steady Demand traced the number to a study published May 5, 2025 by Spanish SEO consultant Natzir Turrado. He used 50 prompts across five Spanish cities and inspected ChatGPT’s raw JSON responses. He found that 60 to 70% of the businesses ChatGPT placed first came from Foursquare. That is a real finding with a narrow scope: five cities, one country, the first-ranked result only, May 2025.
Nine days later, a digital-agency blog was repeating it as a general statement about ChatGPT local results. Over the following year it was rounded up to “70%+,” presented without a source, and in one case paired with a Google and Yelp comparison that did not exist in the original. BrightLocal covered it in July 2025 with a hedge (“Reports suggest”) and links to the sources. Of the eight repetitions Steady Demand found downstream of the original, seven carried no hedge. That is a small sample, and the trace is not exhaustive, so read 87.5% as an illustration and not a statistic.
The study does not say the original Spanish result was wrong when it was measured. It says the broad claim does not describe a large, current U.S. sample. That distinction matters for how we handle numbers on this blog. Every figure here carries its scope, its date and its source, because a stat that loses those three things stops being evidence and starts being folklore. Our own earlier coverage of how AI local search verifies businesses before it recommends them tried to follow that rule, and so should any article you rely on.
Why Is Yelp Behind ChatGPT’s Business Cards?
The likely reason is a licensing deal. Steady Demand cites a non-exclusive Yelp data-licensing agreement reported by Axios on July 23, 2026, about four weeks before its collection window. Axios blocked our automated fetch, so we are relying on Steady Demand’s description of the deal and not on our own reading of the Axios piece. Steady Demand’s own wording is careful: the supported interpretation is a partnership snapshot, not a permanent property of Yelp, Foursquare or ChatGPT.
Here is how the providers compared on the main surface. Shares can overlap because one run can include several providers.
| Provider | Visibly cited | Retrieved | Business-card links |
|---|---|---|---|
| Yelp | 1.04% | 2.08% | 95.83% |
| TripAdvisor | 13.55% | 20.70% | 28.90% |
| OpenTable | 1.70% | 3.58% | 16.88% |
| Resy | 0.00% | 0.00% | 8.96% |
| 0.07% | 0.17% | 4.62% | |
| Foursquare | 0.00% | 0.00% | 0.00% |
Read the columns carefully. The 95.83% is the share of runs where Yelp appeared in the structured grounding attached to ChatGPT’s business cards. Yelp was visibly cited as a link in only 1.04% of runs. So a customer usually does not see the word Yelp. The data under the card is the part that points there. Steady Demand’s measure shows where the card’s data pointed. It does not prove Yelp decides which businesses get named.
The spread across industries matters for our buyers. Yelp’s business-card share ran from 93.75% for hair salons to 97.92% for plumbers. Foursquare was the only one of the ten tracked local-data providers that measured zero in every layer on the main surface. Yelp was not confined to restaurant-style categories.
The researchers also tried the hidden-influence question. They compared the businesses ChatGPT named with ranked business sets from Google and from Foursquare, across 137 market-and-category combinations. The mean overlap was 7.5% with Google and 5.7% with Foursquare, and the difference was statistically significant (Wilcoxon signed-rank p = 0.008). If Foursquare were quietly steering answers, you would expect the opposite. Notice also how low both numbers are. ChatGPT’s picks resembled neither list closely, which fits with other findings that its recommendations vary from run to run. That is why we recommend building a broad, consistent footprint instead of chasing one directory.
Does ChatGPT Recommend Independent Restaurants at All?
Yes, but public data on independents is thin, and the numbers that exist measure different things. Driven Metrics compiled the 2026 research in October 2026, and two figures stand out for restaurants.
SOCi’s 2026 Local Visibility Index covered about 350,000 locations of 2,751 multi-location brands. ChatGPT recommended 5.3% of restaurant locations, against 7.1% for Gemini, 7.6% for Perplexity and 24.3% for appearing in Google’s local 3-Pack. A location counted as recommended only if it landed in the top five of a list of ten. One chain, Culver’s, was recommended 30.0% of the time, so individual brands can sit far from the average. SOCi also noted in its methodology that early patterns suggest models often favor single-location businesses over multi-location brands, which could be good news for independents. Driven Metrics reports that comparable public data for independents was not available as of October 2026.
Local Falcon’s analysis of 189,905 ChatGPT results found that 83% of the restaurants it tracked never appeared in any ChatGPT result in February 2026. On Google, 14% never appeared. That headline is sobering, but it counts a restaurant as visible if it showed up even once across many scans, and it only covers businesses Local Falcon tracks. Driven Metrics is explicit that you cannot compare it with SOCi’s location-level rate. It also flags a 12.3% figure sometimes credited to SOCi that does not appear in the SOCi study it is attributed to.
Put together, the honest summary is this. Most restaurants are not being named by ChatGPT often, and that is true under both methods. The gap between ChatGPT and Google is real. But the studies disagree on size, and none measures independents directly. For an owner in DeKalb or Allen County, the practical takeaway is not a percentage. It is that the listings feeding ChatGPT’s cards need to be accurate, because that is the part you control. For a broader view of how restaurants can handle AI discovery, see our guide to restaurant marketing and AI in 2026.
Accuracy deserves its own mention. In the same SOCi data, ChatGPT’s business details (name, address, phone, hours, website, ratings) were complete 99.9% of the time but accurate only 68.3% of the time. That is close to one in three profiles with at least one error. Wrong hours or a wrong phone number is exactly the kind of error that sends a hungry customer somewhere else.
What Do Diners Do After an AI Names a Restaurant?
They check. Menufy, an online ordering platform for independent restaurants, commissioned a survey from Dynata in September 2026 of 1,000 U.S. consumers aged 18 and older. FSR Magazine reported the results. Menufy sells to the restaurants it surveyed about, so read it as a vendor survey, not neutral research.
The headline numbers favor AI discovery. Sixty percent of respondents said they use AI tools such as ChatGPT, Gemini, Google AI Mode or Perplexity to help decide where to eat. Thirty-five percent said an AI recommendation introduced them to a restaurant they had not heard of and they went on to visit or order. Eighty percent said they would try an independent restaurant they had never heard of if AI recommended it, and 76% said they would be likely to try a local restaurant an AI recommended, compared with 65% for a national chain.
The caution is in the follow-up behavior. Only 2% said they would order from an unfamiliar AI-recommended restaurant without checking another source. Fifty-four percent would review menu information and prices, 48% would check ratings and reviews, and 47% would visit the restaurant’s official website. And 59% said they trust friends or family more than AI.
That sequence is the useful part for a small business. AI may get you into the shortlist, but the listing, the reviews and your website close the deal. It also explains why accuracy matters twice: once in whatever source feeds the AI’s card, and again on the pages a diner opens to verify. Our post on name, address and phone consistency for AI bots in Fort Wayne covers the basic hygiene that makes both checks pass.
What Does This Look Like in a Northeast Indiana Business?
None of these studies sampled Northeast Indiana on its own, and we found no local figures for ChatGPT visibility in Fort Wayne or the surrounding counties. The examples below are illustrations of how the national findings could land. They are not case studies and not client results.
- A family restaurant in Auburn. The owner updated hours on the website and Google Business Profile after a seasonal change but never touched the old Yelp page. If a card pulls from Yelp, a diner sees last year’s closing time. The fix is a ten-minute edit, not a campaign.
- A plumbing or HVAC company in Kendallville. Yelp ran at 97.92% for plumbers in the test. If your Yelp page lists a former phone number or a service area you no longer cover, an AI answer can send a caller to the wrong place.
- A dental or medical practice in Fort Wayne. A claimed Yelp page with correct specialties, hours and a link to the practice website gives an assistant clean data. Keep replies to reviews free of any patient detail, and ask your compliance advisor about the rest.
- A salon or retail shop in DeKalb County. Hair salons showed 93.75% Yelp grounding. A page with current services, photos and prices helps an assistant describe what you do, and helps the shopper who then double-checks.
The pattern in all four is the same: a reasonable listing, nobody assigned to keep it current, and an assistant that repeats whatever it finds. If your business is a Yelp-listed service company, our earlier piece on Yelp’s own AI assistant and Fort Wayne service businesses covers the other side of Yelp’s AI push.
What Should You Fix This Month?
This list is our own, built from the findings above. It is not a guarantee of any ChatGPT result, and nothing here has been tested by Button Block against ChatGPT’s ranking. It is a set of low-cost fixes that also help ordinary search and the customers who check before they call.
The Yelp-First Local Fix List
- Find and claim your Yelp page. Search your business name in Yelp. If a page exists and you have not claimed it, claim it. If you have a duplicate, report it.
- Make Yelp match your Google Business Profile. Same business name, address, phone, hours and website on both. Fix whichever one is out of date.
- Fill in the details assistants repeat. Categories, services, menu or price list, holiday hours, and a few recent photos.
- Keep reviews current and answer them. A steady stream of recent feedback is one of the fundamentals Menufy recommends. Keep replies factual and free of private customer details.
- Restaurants: check TripAdvisor, OpenTable and Resy. Each appeared in business-card data in the test. Update the ones you already use. Do not sign up for new ones just for this.
- Make your website the confirmation page. Diners open the website to verify. Put menu, prices, hours, ordering and dietary options in plain text on the page, not only in a PDF or a photo.
- Do not spend on Foursquare or on Yelp ads for AI reasons. The test gives no support for either.
- Recheck every 90 days. Ask ChatGPT five to ten of the questions your customers ask, several times each, and write down how it describes you. Lists change from run to run, so one answer proves little.
On step 8, Driven Metrics cites SparkToro and Gumshoe research finding less than a 1 in 100 chance that ChatGPT returns the same list of brands in any two of 100 responses to the same prompt. That is why repeated checks matter more than any single screenshot. If you would like help running that audit, our answer engine optimization service and local SEO service start from the same listing and website checks.
How Much Should You Trust These Numbers?
Treat all of it as directional. Here are the limits we would keep in mind.
The key test is one snapshot. Steady Demand collected everything in one 72-hour window in August 2026, four weeks after a competing licensing announcement. It covers U.S., English-language, ChatGPT-only results. The prompts produced generic, non-personalized answers, not what a signed-in user with a history and a location would see. Entity matching between ChatGPT, Google and Foursquare was automated, and recall was not measured. Steady Demand is a third party and we have not reproduced its test.
Card data is not ranking. A Yelp link attached to a card shows where the data pointed. It does not show that Yelp decided who got named, and Yelp was visibly cited in only 1.04% of runs.
The other sources have their own limits. Menufy’s survey was commissioned by a vendor. SOCi measures multi-location brands only. Local Falcon covers only the businesses it tracks. Driven Metrics compiled published research and ran no tests of its own, and we are citing its summary of SOCi and Local Falcon instead of the original pages. Axios and Search Engine Land both blocked our automated fetch, so we did not read their coverage of the Yelp deal directly.
Nothing here is local. We found no Northeast Indiana data and have not invented any. What the studies can do is tell you what to check on your own listings, and that costs nothing.
See What ChatGPT Says About Your Business
Button Block helps Northeast Indiana businesses check how AI assistants describe them, then fix the listings and website details that get it wrong. Start with a conversation about where your business shows up today.
Frequently Asked Questions
- Does ChatGPT use Yelp for local business results?
- In one August 2026 test, yes. Steady Demand ran 2,880 prompts and found Yelp attached to the structured business cards in 95.83% of runs. That measures where the card data pointed, not what a searcher sees: Yelp was visibly cited as a link in only 1.04% of runs. It is a snapshot taken about four weeks after a reported Yelp data-licensing deal, so treat it as current behavior that could change.
- Should I still list my business on Foursquare?
- Foursquare is not a priority for ChatGPT visibility on this evidence. Steady Demand found Foursquare in 0.00% of citations and business-card links on ChatGPT's main surface, and in one of 1,728 runs on a second surface. Foursquare has a real data partnership with OpenAI, and nothing in the test disputes that. A listing you already have is harmless. Time you have not spent yet is better put into Yelp, your Google Business Profile and your own website.
- Where did the 60 to 70% Foursquare figure come from?
- It traces to a study published May 5, 2025 by Spanish SEO consultant Natzir Turrado. He ran 50 prompts across five Spanish cities and found that 60 to 70% of the businesses ChatGPT ranked first came from Foursquare. As other writers repeated it, the Spain-only, first-result-only and May 2025 limits dropped away. The original study may well have been right for its sample. The problem is that the number was later applied to every local result.
- How often does ChatGPT recommend restaurants?
- It depends on how you count, and the studies do not agree. SOCi's 2026 index found ChatGPT recommended 5.3% of restaurant locations belonging to multi-location brands. Local Falcon found 83% of the restaurants it tracked never appeared in any ChatGPT result in February 2026, against 14% on Google. Those use different denominators, so do not compare them directly. We found no public rate for independent, single-location restaurants.
- Will Yelp's role in ChatGPT last?
- Nobody can say. Steady Demand describes its result as a partnership snapshot and warns it should not be treated as a permanent property of Yelp, Foursquare or ChatGPT. The Yelp agreement was reported as non-exclusive, so other sources can still matter. The practical response is to keep your Yelp page accurate without betting everything on it, and to recheck how ChatGPT describes your business every quarter.
- Do I need to buy Yelp ads to show up in ChatGPT?
- Nothing we read supports that. None of the sources tests whether paid Yelp advertising changes what ChatGPT returns. The Steady Demand finding concerns the free business listing data attached to the cards. Claim and complete your free Yelp page first. If you want Yelp ads for the traffic they send you directly, judge them on that basis, not on a promise about AI visibility.
- Can Button Block help us check how ChatGPT sees our business?
- Yes. Button Block works with Northeast Indiana businesses on local SEO and answer engine optimization. A good first step is the one in this post: ask ChatGPT the questions your customers ask, record what it says about you across several runs, and fix the listings and website details it gets wrong. You can start with a conversation on our contact page.
Sources & Further Reading
- Steady Demand: ChatGPT’s Local Results Aren’t Coming From Foursquare: 2,880-prompt test, August 2026 (all Yelp, Foursquare, provider and overlap figures)
- Driven Metrics: ChatGPT Local Business Recommendation Rate 2026: compilation of SOCi, Local Falcon, SparkToro and other 2026 research
- FSR Magazine: How AI Is Influencing Independent Restaurant Dining Decisions: Menufy and Dynata consumer survey, September 2026
- Local Falcon (as cited by Driven Metrics): The AI Visibility Crisis: Why 83 Percent of Restaurants Don’t Exist in ChatGPT: February 2026 restaurant analysis
- Axios (as cited by Steady Demand): Yelp reviews and ChatGPT partnership report, July 23, 2026 (we could not open this page directly)