Does Location Matter with AI Search?
AI search does not give every buyer the same answer. A new study of 31,200 answers shows that ChatGPT changes its sources city by city and Gemini changes them country by country. Here is what that means for how you track and win AI visibility.
In short
Yes. Location matters a lot for local questions. ChatGPT changes its sources and its recommended businesses city by city. Gemini changes them country by country. So an AI visibility report from one place shows one market only. If you sell in more than one city, test and build your visibility in each city.
Our view: one test location gives you the wrong picture
Most AI visibility checks ask a set of questions from one place. They show one answer and call it your visibility. We think that number misleads local businesses, multi-location brands and the agencies that report to them.
A new study now gives the data. In Location, Location, Generation: Where You Ask Matters for Generative Engine Optimization, Mahe Chen and Nick Koudas of the University of Toronto asked the same questions from different places. They measured how much the sources and the businesses changed.
The result is clear. For a near me question, two US cities shared only 8% of the businesses that ChatGPT recommended. Two runs in the same city shared 32%. Your buyer in Sacramento and your buyer in San Jose do not see the same answer.
We covered the study first in A/GEO Research Weekly, Issue 2. This post goes deeper into what it means for your business.
What the study tested
The researchers wrote 300 questions in ten commercial topics, such as home services, health care, finance, dining, cars and public services. No question named a place. They put the questions into three groups of 100:
- Near me: the buyer wants something close by, such as "best plumber near me."
- Depends on place: the answer changes with the place, such as sales tax on a car, laws or local providers.
- Not local: the answer is the same everywhere, such as "best robot vacuum."
They tested two ways that location gets into an AI answer:
- Stated location: the app tells the AI model where the user is. ChatGPT got 19 cities: 15 US cities plus London, Toronto, Sydney and Dublin.
- Network location: the AI model sees where the internet connection comes from. ChatGPT and Gemini got requests from Canada and from VPN exits in New York, Los Angeles, London and Sydney.
Each request ran twice, so the team could separate normal run-to-run noise from a real location effect. In total, they collected 31,200 answers.
The main measure is the localization index. It takes the overlap of cited sites between two identical requests and subtracts the overlap between two places. A score of 0 means that location has no effect beyond noise. A higher score means that location changes more of the cited sites.
Location changes the sources for local questions
For questions that depend on place, ChatGPT changed its cited sites by 0.43 between two countries. For not-local questions, the change was 0.03, close to noise. ChatGPT uses location only when the question needs it.
Gemini acts differently. Its sources changed by 0.23 between two countries even for not-local questions. A search for a robot vacuum from London and from Sydney gets different sources.
| Comparison | Near me | Depends on place | Not local |
|---|---|---|---|
| ChatGPT, stated city: two US cities | 0.14 | 0.17 | 0.00 |
| ChatGPT, stated city: two countries | 0.28 | 0.43 | 0.03 |
| Gemini, network: two countries | 0.43 | 0.40 | 0.23 |
| Gemini, network: US East vs West | 0.00 | about 0.02 | about 0.03 |
| ChatGPT, network: two countries | 0.00 | 0.01 | 0.01 |
Source: Chen and Koudas (2026), Figure 1. The Gemini US East vs West values for "depends on place" and "not local" are read from the figure. All other values come from the paper's text.
A stated location also changes whether ChatGPT cites sources at all. With a city, it cited sources in 64% to 73% of answers. Without a city, it cited sources in 46% of answers. Without a location, ChatGPT asked for a ZIP code or city on 63% of near me questions.
ChatGPT reads the city. Gemini reads the country.
This is the finding that changes how you should test. The two engines take location from different places.
ChatGPT follows the stated city and ignores the network. Ask for "credit unions in my area" with Boise as the city. ChatGPT searches for "credit unions Boise Idaho area branches near Boise Idaho." Change the city to London and it searches for "credit unions near London England United Kingdom." It wrote the place into 100% of its searches for near me questions.
The search system behind ChatGPT also adds location by itself. When neither search named a place, the sources still changed (index 0.17). For a user in Boise, ChatGPT cited Oregon State University's extension service on when to plant tomatoes.
Distance does not matter to ChatGPT. Two US cities less than 1,000 km apart scored 0.13. Two cities more than 3,000 km apart scored 0.14. To ChatGPT, the next town is a different market.
Gemini follows the network country and ignores the stated location. Between New York and Los Angeles, its index was 0.00. Its searches named the country in 56% of near me answers, but never a city. Ask "which store has the cheapest groceries" and Gemini looks at UK stores from a UK network and Australian stores from an Australian network.
What this means: a test that sets only a city measures ChatGPT for that city but Gemini for the country the test runs from. A test that sets neither measures a general US user in ChatGPT and the test's own country in Gemini. Ask your tool or agency which way it sets location for each engine.
Some industries depend on place more than others
Location does not matter the same amount in every topic. For place-dependent questions, public services scored 0.57, because answers cite each state's government sites. Telecom and utilities scored 0.02, because answers cite national sources such as the FCC, national carriers and comparison sites.
| Topic | Near me | Depends on place |
|---|---|---|
| Public services | 0.16 | 0.57 |
| Cars, transport | 0.08 | 0.27 |
| Food, dining | 0.11 | 0.24 |
| Personal finance | 0.14 | 0.19 |
| Health care | 0.12 | 0.17 |
| Home services | 0.10 | 0.17 |
| Insurance | 0.12 | 0.09 |
| Travel, leisure | 0.31 | 0.06 |
| Retail | 0.15 | 0.04 |
| Telecom, utilities | 0.12 | 0.02 |
Source: Chen and Koudas (2026), Figure 2. Public services (0.57) and telecom (0.02) for place-dependent questions come from the text. The other values are read from the figure, accurate to about 0.01. Each value rests on ten questions, so use it as a direction, not an exact score.
Our read: if you sell a service people search for near them, such as dining, car services, health care or home services, location shapes your AI visibility. If your buyers ask about rules, taxes or providers in their state, it shapes it even more. National brands in telecom or retail face less change between US cities, but they still face it between countries.
Each city gets a different list of businesses
Sources are only half of the story. The businesses that AI recommends change with them. A ChatGPT answer to a near me question recommended 4.3 businesses, brands or products on average. Two identical requests from the same city shared 23% of them. Requests from two different cities shared 4%.
Here are the first three credit unions that ChatGPT recommended in both runs, for each stated city:
| Stated city | ChatGPT: "credit unions in my area" |
|---|---|
| New York | New York University FCU, Neighborhood Trust FCU, Brooklyn Cooperative FCU |
| Boise | Idaho Central CU, CapEd FCU, Connections CU |
| Miami | Educational FCU, University CU, Miami Postal Service CU |
| London | London Mutual CU, London Capital CU, Your Community Bank |
Gemini changes its picks by country, even when the question names no place. For "which store has the cheapest groceries," it recommended No Frills, Walmart and Food Basics in Canada. In the UK, it recommended Aldi, Lidl and Tesco. In Australia, it recommended ALDI, Coles and Woolworths. New York and Los Angeles got the same top two: Aldi and Walmart.
Our read: you do not compete with every business in your category. You compete with the businesses that AI recommends in each city you serve. The list in one city tells you little about the list in the next one.
How many cities should you track?
The study tested how much a single-city check misses. For near me questions, 96% of the businesses that ChatGPT recommended appeared in only one of the 15 US cities. Move the slider to see how much of the full list you see as you add cities.
| Cities tracked | Near me | Depends on place | Not local |
|---|---|---|---|
| 1 | 9% | 16% | 19% |
| 3 | 23% | 33% | 39% |
| 5 | 36% | 47% | 53% |
| 10 | 69% | 76% | 79% |
| 15 | 100% | 100% | 100% |
Source: Chen and Koudas (2026), Figure 3. Values for 1, 5 and 10 cities on the near me and not local lines come from the text. The other points are read from the figure, accurate to about 1 point.
The near me line does not level off. Each new city brings mostly new businesses. Even the not-local line, which location does not affect, starts at only 19%. That gap comes from run-to-run variation. So a good check needs two things: repeated runs, and for local questions, many places.
What this means for your business
Here is how we read the study for the businesses we work with:
- Local service businesses: Each city or town you serve is its own AI search market. A strong answer in your home city does not carry to the next suburb.
- Multi-location brands: Track each location's market on its own. Give each location a page that names its city and its services, so AI search finds a page for each place.
- Brands in more than one country: Check Gemini from a network in each country, even for questions that name no place.
- Agencies and in-house teams: Put the test city and the network country on every AI visibility report. Compare month to month, or engine to engine, only with the same locations.
The study also shows where local visibility comes from. In US cities, 79% of ChatGPT's near me answers cited a source that names the city or state. Those sources are local directories, review sites and the sites of local institutions. National rankings do not decide local AI answers. Pages about your place do.
How to run a three-city test this week
- List your markets. Write down each city you serve. For ChatGPT, a nearby city does not stand in for another one.
- Pick five money prompts. Choose near me and place-dependent questions that buyers ask before they buy, such as "best HVAC company near me."
- Set the location on the right channel. For ChatGPT, state the city. For Gemini, send the request from a network in the target country. A location setting alone does not move Gemini.
- Run each prompt twice per city. Record the cited sites and the recommended businesses. Two runs show you what is noise and what is location.
- Find the local pages AI cites. Look for the directories, review sites and local institutions in the results. Get listed and reviewed there. Make sure your own pages name each city you serve.
- Repeat with the same cities each month. Keep the locations fixed, so a change in the result is a real change.
Limits of the study
- The team used the developer APIs and small models. The consumer apps can act differently. A check with the full-size ChatGPT model on 720 answers showed equal or stronger localization (0.26 against 0.23).
- VPN exits are data-center addresses. Gemini can localize more finely on home connections.
- The questions are in English, and the authors wrote them.
- Each topic result rests on ten questions.
- The results describe the engines in October 2026. The engines change often.
Common questions
Does location matter with AI search?
Yes, for local questions. In a University of Toronto study of 31,200 answers, the user's location changed most of the sources that ChatGPT and Gemini cited for near me questions. It also changed the businesses that they recommended.
How does ChatGPT know where a user is?
In the study, ChatGPT used the location that the app stated, down to the city. It ignored the network location. With no stated location, it asked for a ZIP code or city on 63% of near me questions.
How does Gemini know where a user is?
In the study, Gemini ignored the stated location and used the country of the network. It did not separate New York from Los Angeles. It did change its sources between countries, even for questions that do not depend on place.
How many cities should I track for AI visibility?
Track every city that you serve. In the study, one city showed 9% of the businesses that ChatGPT recommended across 15 US cities for near me questions. Five cities showed 36%, and ten cities showed 69%.
Can my AI visibility tool show a wrong result?
Yes, if it checks from one place. A tool that sets only a location setting measures ChatGPT for that city but Gemini for the country that the tool runs from. A tool that sets neither measures a general US user in ChatGPT and its own country in Gemini.
Sources
- Chen, M., and Koudas, N. (2026). Location, Location, Generation: Where You Ask Matters for Generative Engine Optimization. University of Toronto. arXiv 2610.24081. Figures 1 to 3 and Tables 1 to 4.
- A/GEO Research Weekly, Issue 2: Where You Ask Changes AI Answers, October 19, 2026.