Getting recommended by ChatGPT: read its sources before paying

A customer in Porto who needs a physiotherapist no longer has to open Google Maps. They can type "who is a good physio near Boavista?" into ChatGPT and get three names with a sentence on each. BrightLocal, a local-search research firm, found in its 2026 Local Consumer Review Survey that 45% of US consumers had used an AI tool for local business recommendations in the past year, up from 6% the year before. So owners now ask how you get recommended by ChatGPT, and the guides that answer disagree with each other. One says claim your Foursquare listing, one says Yelp, one says your Google profile is all that matters. Before you pay anyone to fix a listing, it helps to know which of those claims has been measured, and what the measurements found ChatGPT reading.

TL;DR: Nobody outside OpenAI knows which directory feeds ChatGPT's local answers, and the popular Foursquare claim found no support in a 2,879-query test in August 2026. In the one study that counted its local sources, business websites led. To get recommended by ChatGPT, ask it your customer's question, read its sources, and fix your own site first.

How does ChatGPT decide which local businesses to recommend?

OpenAI has published almost nothing about how ChatGPT picks local businesses. Its ChatGPT search help page says the assistant may share a user's location with "trusted third-party providers" to supply nearby business listings, and it names none of them. It also decides for itself whether to search the web at all, so some answers come from what the model already holds and some from a live search.

What happens after the businesses are found is even less documented. The only study that looked inside ChatGPT's raw response data, by Spanish SEO consultant Natzir Turrado in May 2025, found the businesses listed in the order the data providers answered, with no sign of sorting by rating. That was a small sample, taken more than a year before ChatGPT started using Yelp's data.

So any guide that lists the exact "ranking factors" for ChatGPT's local answers is describing a guess. That matters when someone quotes you a monthly fee to optimise them.

Does ChatGPT use Foursquare, Yelp or Google reviews?

The claim repeated most often is that 60 to 70% of ChatGPT's local results come from Foursquare. It traces back to Turrado's study: 50 prompts, five Spanish cities, the first-ranked result only. In August 2026 Ben Fisher of Steady Demand, a US local-search consultancy, ran 2,879 successful ChatGPT queries across 12 US cities and 12 business types, weighted towards restaurants, bars and hotels where Foursquare should be strongest. Foursquare's sites appeared in 0.00% of the sources ChatGPT cited and in none of its business cards, the small panels with a rating and a link that sit inside an answer. Fisher also traced eight later articles that repeated the 60 to 70% figure; seven had dropped every caveat.

Yelp is the one source confirmed by both companies. Yelp says its reviews, ratings, photos and business details already appear inside ChatGPT, and on 10 August 2026 it added table bookings for restaurants in the US and Canada. In Fisher's test, Yelp data sat behind ChatGPT's business cards in 95.83% of runs.

Google's role is disputed. BrightLocal's 2026 survey commentary says ChatGPT "can't see inside Google's walled garden of reviews", meaning the reviews on Google Maps. Fisher found the businesses ChatGPT named overlapped slightly more with Google's results (7.5%) than with Foursquare's (5.7%), and both overlaps are small: ChatGPT's list rarely matched either one. Facts from a Google profile also get copied onto other websites, so they can reach the model that way.

For a business in Portugal or anywhere else in Europe, every large test is US, English-language and dated, and the Yelp booking deal covers the US and Canada. There is no confirmed directory to prioritise, which leaves the sources you can check yourself.

Which sources does ChatGPT read for local searches?

Mostly business websites, in the one study that counted. In November 2024 BrightLocal ran 800 local searches in ChatGPT across 20 business types and 20 US cities and recorded the first ten sources listed under each answer. Business websites made up 58% of them, mentions on other sites (Wikipedia, local press, "best of" roundups) 27%, and directories 15%. For plain searches such as "coffee shop" rather than "best coffee shop", business websites were 72% of the sources. Yelp, Facebook and Google Maps did not appear as directory sources in that sample, which was taken before the Yelp deal.

That study is almost two years old, so the exact shares will have moved. When ChatGPT searched, it mostly read the sites of the businesses it was choosing between.

A directory decides what it shows about you. Your website says what you write on it, and it is the page an assistant can quote when it explains why it picked you.

We build websites for a living, so weigh our view accordingly. The website is still the input every owner controls, and it does its job with human visitors whether or not an assistant ever names you.

How to check whether you are recommended by ChatGPT

The check takes about 30 minutes and needs no tool.

1. Write down the question a real customer would ask, in their words and their language, with the place in it: "physio near Boavista that does sports injuries", not "best physiotherapy". A logged-out session may not know where you are, so the place has to be in the question.
2. Open ChatGPT in a private window, logged out, and ask it ten times. Record every business it names and every source it links. One answer means very little, because the list changes from run to run.
3. Ask the same question for a neighbouring town or district. Coverage often thins away from city centres, and you want to know whether the gap is you or the area.
4. Ask for your business by name. Read which pages it cites and whether the facts are right: services, area, opening hours, who you work with.
5. If a business card appears, note which platform supplied it.

Step 4 is where the fixable problems show up. If ChatGPT describes your business from a two-year-old directory entry, or cannot say what you actually do, the page it should have read is your own.

What should your website say so an assistant can repeat it?

An assistant can only repeat what is written down in plain text. A homepage that says "quality care, tailored to you" over a photo gives it nothing to quote. A page that says "sports injury physiotherapy in Boavista, Porto, for runners and weekend footballers, Monday to Saturday" gives it a sentence it can lift whole.

For the physio in Porto, that means:

• A page called "Sports injury physiotherapy", rather than a "Services" page with six treatments in a carousel.
• Boavista, Foz and Matosinhos named in the text as well as shown on the map.
• The answers a patient checks before calling: hours, whether they need a doctor's referral, whether you see them in English, and a price range if you publish one.
• Proof that names something: the professional body you belong to, the clubs whose players you treat. "Trusted by hundreds" gives an assistant nothing specific to repeat.

We do this on our own site. sharphaw.com publishes a plain-text summary file and a plain-text pricing file, and the "ask an assistant" links in our footer ask ChatGPT and the others to answer from those pages rather than from what the model remembers. That will not make an assistant recommend us, and we have no way to prove it changes what they say. When one does describe SharpHaw, it has our own wording to work from.

The open question is whether being named by ChatGPT brings calls or bookings. Nobody has published a measurement of it yet. On a conversion-first website the service and area pages are part of the build, because human visitors need them first.

Frequently asked questions

Can I pay to be listed in ChatGPT's local recommendations?

Not through any route OpenAI documents. Zarla, a website builder that runs these checks for its customers, reports finding no ad product or paid placement in ChatGPT's local recommendations. Treat any offer that guarantees a spot in those answers as a claim to test before you pay.

Does ChatGPT recommend businesses in small towns?

Less reliably. Zarla reports that city-level questions return lists while suburb-level questions often name nobody. If ChatGPT names no one for your village, that is its coverage today, and the question to ask is whether it names you for the nearest town.

How often should I repeat the check?

Every few months, and after any change to your site. The Yelp licensing deal was reported in July 2026, bookings followed in August, and the sources shift with each new deal, so a result from last spring tells you little. Keep the questions and the date of each run so you can compare like with like.

Start with the page you control

The directory theories will keep changing as OpenAI signs new data deals, and much of what is sold as ChatGPT optimisation rests on a Foursquare figure that found no support in its first large test. Two things hold either way. Ask ChatGPT your customer's question yourself before you change anything, and make your own website say what you do, where and for whom, in sentences an assistant can quote.

If you want a second pair of eyes on what your site tells an assistant today, book a 30-minute call and we will go through it together. If the site itself needs rebuilding, the Foundation plan covers the website and its upkeep, and every fee is on the plans page.