Guides

How Does ChatGPT Decide Which Local Businesses to Recommend?

OpenAI doesn't publish a formula. What its documentation says, what studies of thousands of AI answers found, and what remains unknown, ranked by how strong the evidence is.

November 11, 2026 · 5 min read

If you've ever wondered why ChatGPT named your competitor and not you, you've run into a hard fact: nobody outside OpenAI knows exactly how ChatGPT chooses local businesses, and OpenAI hasn't published a formula.

What we do have is OpenAI's own documentation, and a growing set of studies that counted what ChatGPT cited and which businesses it named. This guide ranks what's known by how strong the evidence is, so you can tell a solid lead from a rumor. Almost every study here is observational and many were run by companies that sell related services, so read the conclusions as patterns, not proof.

What OpenAI says

OpenAI's public pages say a few things:

That is the whole official picture. Everything else is inferred from outside studies.

What the studies found

Strong evidence: what ChatGPT reads

It reads websites and directories more than review sites. Yext's analysis of 155.5 million AI citations (Jan to Mar 2026) found about 49% of OpenAI's citations went to business websites and about 36% to listings and directories, while reviews and social made up only about 2.6% (Yext). Yext sells listings and mostly studied brands, not one-truck contractors.

For home services, it leans on specific directories. A Cheers study of 34,742 answers across 100 metros (September 2026) found ChatGPT cited Expertise.com about 32%, Angi/HomeAdvisor about 29%, BBB about 15%, and Google 0.1% (Cheers). In our own November 2026 study of 390 Texas markets, ChatGPT cited Angi in 78.5% of markets and a ranking site, ContractorsRanked, in 71.8%, and cited Yelp and Facebook in none. That is ChatGPT's search through OpenAI's API, one day, Texas only.

Moderate evidence: which businesses get named

Reviews, especially count and freshness, go with being named. A November 2026 study of about 2,900 answers (Revfer) found review count was the strongest predictor of being named, and ChatGPT named about 20% of businesses whose newest review was under a week old versus under 2% when it was a year old or more (Revfer). It's observational: busy businesses naturally have both.

Rating works more like a filter. SOCi found recommended locations averaged about 4.2 to 4.3 stars on ChatGPT and Perplexity (SOCi).

Weaker evidence

What nobody knows

What you can do with this

Focus on the things that are both plausible and cheap:

  1. Let the crawler in. See how to allow OAI-SearchBot.
  2. Be on the sites ChatGPT cites for your trade, with accurate details. See the directories ChatGPT reads and the "best of" sites.
  3. Keep reviews fresh and honest. See how many reviews and how recent.
  4. Make your website clear: a page per service, the cities you serve, license and insurance details, real FAQs. See service pages and FAQs that answer "who should I call?".
  5. Check results monthly, asking each question several times. See the five-minute test.

How to read the next study you see

New studies about AI search appear every week. A quick way to judge them:

  1. Who ran it? A company that sells listings, reviews or AI-visibility services has a reason to find that those things matter. That doesn't make the data wrong, but read it with that in mind.
  2. How many answers, which tools, which dates? Thousands of answers across many cities beat a few dozen. AI tools change quickly, so a study from late 2024 may not describe today.
  3. Observational or an experiment? Most are observational: they show what went with being named, not what caused it. The only controlled test we found was on schema markup, and it found no effect.
  4. Does it cover your kind of business? Studies of restaurants, dentists or national brands may not describe a roofer or plumber. The closest we have to home services are the Cheers study (34,742 answers) and our Texas study (2,238 answers).
  5. Does it name the number it measured? Be cautious of percentages with no stated method.

What to ignore

FAQ

Does ChatGPT use Google's rankings? We found no evidence that it does. Cheers saw ChatGPT cite Google just 0.1% of the time for home services, and SOCi found that 35.9% of locations appeared in Google's local 3-pack but only 1.2% were recommended by ChatGPT. Being strong on Google doesn't mean being named by ChatGPT.

Can I pay to be recommended? OpenAI sells ads, which are labeled and separate. See ChatGPT ads for contractors. Nothing we found says paying changes organic recommendations.

Why did my competitor get named? Often because of better coverage on the sites ChatGPT reads. See why ChatGPT recommends your competitor.


See what ChatGPT reads about you. Run a free scan at localpickai.com. You'll see whether ChatGPT names you, who it names instead, which websites it read, and what to fix first. Free, no signup, about a minute.

Check your own business

Free 60-second scan: whether ChatGPT recommends you, who it picks instead, and exactly what to fix.

Free. No signup. Results in about a minute.

Preview of the AI Visibility Checklist for Texas Contractors

Free: the AI Visibility Checklist for Texas Contractors

12 things ChatGPT checks before it recommends a roofer, plumber or HVAC company, each with a 2-minute check you can do yourself.

We’ll email you the checklist and occasional tips. Unsubscribe anytime.

More guides

Free scan