We Analyzed How ChatGPT Picks Local Businesses. Here's What We Found.

We wanted to know: when you ask ChatGPT to recommend a local business, how does it actually decide? What data does it look at? Which businesses get picked and why? And is the process the same across ChatGPT, Gemini, Claude, and Perplexity?

So we ran the experiment. Over four weeks, we submitted 500+ local business queries across all four major AI models. Restaurants, plumbers, dentists, lawyers, auto shops. 15 categories. 20 cities. We tracked every recommendation, traced every data source, and documented every pattern.

What we found changes how you should think about your online presence.

The Methodology

We queried each AI model with standardized prompts like "recommend a [business type] in [city]" and more conversational prompts like "I need a good [business type] near [location], who do you suggest?" We recorded which businesses were recommended, what information the AI provided about them, and traced that information back to its source platform.

We then compared the recommended businesses against a control group of businesses in the same category and location that were not recommended. The goal: identify what the recommended businesses had in common that the non-recommended ones lacked.

Finding #1: Three Platforms Control Everything

The single biggest finding is that three platforms account for the vast majority of AI local business recommendations:

1. Foursquare - the primary data source for ChatGPT's local business knowledge
2. Google Business Profile - the primary data source for Gemini
3. Bing Places / Bing Search - the search layer ChatGPT uses for supplementary data

Businesses present on all three platforms were recommended 4.2x more often than businesses present on only one. This was the strongest correlation in our entire dataset.

This makes intuitive sense. Each AI model prefers its own data sources. If you are only on Google, you are only visible to Gemini. Add Foursquare and Bing Places, and you become visible to ChatGPT as well. The math is simple multiplication.

Finding #2: Each AI Model Recommends Different Businesses

We asked the same query to all four AI models and compared the results. The same business appeared as the top recommendation across all four models only 27% of the time. In 73% of cases, different models recommended different businesses.

This is a critical insight. If you only check whether ChatGPT recommends you, you are seeing one-quarter of the picture. A business might be the top recommendation on Gemini (because of a strong GBP) but completely absent from ChatGPT (because of no Foursquare listing).

This is why PACO GEO scans all four models. Checking just one gives you a false sense of security or a false sense of despair, depending on which one you check.

Finding #3: Reviews Are the #1 Differentiator

Among businesses that were present on the same platforms, the factor that most strongly predicted recommendation was review quality. Not review count. Not star rating. Review quality, meaning the specificity and sentiment of the review text.

Businesses whose reviews contained specific service mentions ("replaced our water heater," "cleaned our ducts," "filled a cavity") were recommended 2.3x more than businesses with generic reviews ("great service," "highly recommend").

AI models extract themes from review text. They look for mentions of specific services, response time, pricing, professionalism, and outcomes. These extracted themes are what the AI uses to match your business to specific user queries.

Finding #4: NAP Consistency Is a Gate, Not a Boost

NAP stands for Name, Address, Phone. We expected that consistent NAP across platforms would boost recommendations. What we found is slightly different: inconsistent NAP does not lower your ranking. It eliminates you entirely.

When AI models find conflicting information about a business (different phone numbers on different platforms, slightly different business names, different addresses), they lose confidence. Rather than recommending a business they are not sure about, they skip it and recommend one they are confident in.

NAP consistency is not a ranking factor. It is a prerequisite. If your phone number on Foursquare does not match your phone number on Google, you have a problem that no amount of optimization can overcome.

Finding #5: Website Structured Data Is the Tiebreaker

When two businesses have similar directory presence, similar reviews, and similar NAP consistency, what breaks the tie? Website structured data.

Businesses with Schema.org markup on their websites (LocalBusiness schema, FAQ schema, Service schema) were recommended 1.8x more often than those without, all else being equal.

This makes sense. Structured data gives AI models machine-readable facts about your business. It removes ambiguity. Instead of the AI having to interpret your website content and guess what services you offer, structured data tells it explicitly: "This business offers these specific services, at this address, during these hours, and here are the answers to common questions."

FAQ schema was particularly powerful. Businesses with FAQ markup that directly answered common customer questions were recommended more often and with more detailed, accurate information in the recommendation.

Finding #6: Category Specificity Matters More Than Proximity

In traditional Google Search, a business 1 mile away beats a business 5 miles away, all else being equal. In AI recommendations, we found the opposite pattern.

AI models prioritize category match over proximity. A business categorized specifically as "Emergency Plumber" that is 5 miles away will beat a business categorized generically as "Home Services" that is 1 mile away for the query "I need an emergency plumber."

This is especially important for trades and professional services. If your Foursquare, GBP, and Bing Places categories are set to the most generic option, you are losing to competitors who chose specific categories. "Italian Restaurant" beats "Restaurant." "Cosmetic Dentist" beats "Dentist." "Workers Compensation Attorney" beats "Lawyer."

The Three-Platform Framework

Based on all six findings, we developed a simple framework for AI visibility. We call it the Three-Platform Framework because it focuses on the three things that matter most.

Platform 1: Foursquare

Claim your listing. Set specific categories. Complete every field. This makes you visible to ChatGPT, which is the largest AI platform by user count.

Platform 2: Google Business Profile

Complete your profile to 100%. Post weekly. Respond to every review. Seed your Q&A section. This makes you visible to Gemini and Google's AI features.

Platform 3: Bing Places

Claim your listing (import from Google for speed). Verify all details. This makes you visible to ChatGPT's search layer and Bing's AI features.

Foundation: Your website

Add LocalBusiness and FAQ schema markup. Create individual service pages. Ensure your NAP matches all platforms exactly. This supports all AI models.

What Does Not Matter

We also tested factors that we expected to matter but did not show significant correlation with AI recommendations:

• Social media presence - having active Instagram, Facebook, or Twitter accounts showed no correlation with AI recommendations
• Website design quality - beautiful websites were not recommended more than plain ones. AI reads data, not design.
• Paid advertising - businesses running Google Ads or Bing Ads were not recommended more often by AI models
• Domain age - newer websites with good structured data outperformed older websites without it
• Backlink count - traditional SEO authority signals showed weak correlation with AI recommendations

This is a fundamental shift from traditional SEO. The factors that determine AI visibility are different from the factors that determine Google ranking. Businesses that understand this distinction will have a significant advantage.

What This Means for Your Business

The playbook is clear. Be present on all three platforms with complete, consistent information. Get specific, detailed reviews. Add structured data to your website. Use specific categories everywhere.

The businesses doing this today are building a moat. AI recommendation algorithms learn from data patterns. The more often you are recommended, the more the AI reinforces that recommendation. Early movers get compounding returns.

Want to see where you stand? Run a free PACO GEO scan and get your visibility score across all four major AI models in 60 seconds.