# Does ChatGPT Give Different Answers by Country? We Tested It


Yes. When we asked ChatGPT "What number should I call in an emergency?" from the United States, Brazil, and Japan, we got three different answers, each correct for its own country. OpenAI says that when ChatGPT searches the web, it may use an approximate location based on your IP address, and Google says it estimates "the general area that you're searching from" on every search. The same thing happens to AI agents that read the web for you, which is why the common belief that they don't need geotargeted web access is wrong. An agent that ignores location isn't neutral. It sees the answer for wherever its request arrives from, and geotargeting is how you choose that place on purpose.

> **Key Takeaways**
>
> - "When you search on Google, Google will always estimate the general area that you're searching from" ([Google Search Help](https://support.google.com/websearch/answer/179386?hl=en)).
> - ChatGPT "may use an approximate location based on your IP address to provide relevant local results" ([OpenAI Help Center](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt)).
> - Researchers logging prices from different vantage points found several retailers whose prices for the same product varied 10% to 30%, with isolated cases up to 2x ([Mikians et al.](https://arxiv.org/abs/1307.4531), 2013).
> - In our own test on October 10, 2026, the same ChatGPT prompt sent from the U.S., Brazil, and Japan returned three different emergency numbers: 911; 190, 192, and 193; and 110 and 119. Anyone can rerun it on Massive's public [AI geo playground](https://www.joinmassive.com/playgrounds/ai-geo).
> - Location doesn't matter for every task. It matters whenever the answer is local: search results, AI answers, prices, stock, ads, or legal availability.

## We asked ChatGPT the same question from three countries

To see this first-hand, we used Massive's public [AI geo playground](https://www.joinmassive.com/playgrounds/ai-geo) on October 10, 2026. It sends one prompt to ChatGPT through Massive's AI chat endpoint from three countries at once, and anyone can use it without an account. We picked a question with a correct answer that depends entirely on place: "What number should I call in an emergency?"

| Asked from | What ChatGPT answered |
|------------|-----------------------|
| United States | 911 for medical, fire, and police |
| Brazil | 190 for police, 192 for an ambulance, 193 for the fire department |
| Japan | 119 for an ambulance or fire, 110 for police |

![The same prompt, "What number should I call in an emergency?", sent to ChatGPT from three countries on October 10, 2026. From the United States the answer was 911. From Brazil it was 190 for police, 192 for an ambulance, and 193 for the fire department. From Japan it was 119 for an ambulance or fire and 110 for police.](assets/same-prompt-three-countries.png "Same prompt, three countries, three answers.")

Each answer was right for its country, and none of them would help a user in the other two. An agent that fetched this answer from a single data center would hand every user the same number, no matter where they were.

A second, looser prompt showed the same pattern. When we asked for "a typical price for a cup of coffee at a cafe," the U.S. answer quoted dollars, the Brazil answer added prices in reais, and the Japan answer added a figure in yen. The prompt never mentioned a country. The location of the request supplied it.

You can rerun both prompts yourself on the playground, or swap in any of its other countries. Answers can change from run to run, and one run per country shows that the effect exists, not how often it happens. Even so, the location effect is easy to see.

## Does ChatGPT use your location?

Yes, in two ways. OpenAI's help center says ChatGPT "may use your IP address to estimate your general location, such as your country, state, or city," and gives an example: if ChatGPT determines from your IP address that you are in the San Francisco area, it "may rewrite your prompt into the search query 'top restaurants San Francisco'" ([OpenAI Help Center](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt)). It adds that "a VPN or network location may affect the approximate location inferred from your IP address."

Users can also share their device's location, which OpenAI says is "optional and off by default" ([OpenAI Help Center](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt)). Your agent has neither a phone nor a settings page, so for it, the IP address is the location. The question comes up often. In a 2024 [OpenAI community thread](https://community.openai.com/t/does-chatgpt-make-different-answers-based-on-country-or-places/699403), a user asked whether ChatGPT answers differently by country after seeing different results in Japan and Australia. The first reply said, "As far as I know, there are no differences," while later posters described regional differences they couldn't explain. Our test above settles one clear case: a question with a single correct answer per country got a different answer in each.

## Where does the myth come from?

From a reasonable shortcut. Developers test an agent from a laptop or a cloud region, the pages look right, and the agent ships. Many of the pages an agent reads, such as API docs, encyclopedia entries, and standards, really are the same everywhere. The trouble is that the test location quietly becomes the agent's location for every user, in every market, forever.

A cloud server has a location too. Sites estimate it from the IP address, the same way they estimate a person's. So the question was never "does my agent have a location?" It always has one. The question is whether that location is the one your user needs.

## How do sites decide what location to serve?

Mostly from the IP address, through [IP geolocation](https://www.joinmassive.com/glossary/what-is-ip-geolocation). Sites can also read browser language settings and signed-in account preferences, but the IP is the signal every request carries. Google puts it plainly: "IP addresses are roughly based on geography. This means that any website you use, including google.com, may get some information about your general area" ([Google Search Help](https://support.google.com/websearch/answer/179386?hl=en)). The same page notes that "your location helps to show you nearby results, even if you didn't include a location in your search."

![How a page gets localized. The agent's request arrives from an IP address. The site estimates a country, region, or city from that IP. It then serves the matching version: local results, prices and stock, language, ads, and availability. A cloud server's IP produces the version for the data center's location.](assets/how-location-is-chosen.png "Every request carries a location, whether you set it or not.")

## What changes by location?

More than people expect. These are the categories where the same URL or the same question can come back different:

| What the agent reads | What can change by location |
|----------------------|-----------------------------|
| Search results | Local businesses, rankings, People Also Ask |
| AI Overviews and AI chat answers | Whether an AI answer appears, its language, its sources |
| Product pages | Price, currency, stock, delivery options |
| Ads | Which advertisers appear, and which offers |
| Content and media | Whether the page is available at all |

Google AI Overviews alone vary by market. Google's May 2025 announcement was titled "AI Overviews expand to over 200 countries and territories, more than 40 languages" ([Google](https://blog.google/products-and-platforms/products/search/ai-overview-expansion-may-2025-update/), May 20, 2025). An agent that checks how a brand appears in AI Overviews from one country is measuring one market. Our guide to [capturing Google AI Overviews and People Also Ask](https://www.joinmassive.com/blog/search-ai-overviews-paa) shows how to pull them from a chosen location. If the page your agent needs is withheld in some countries, that is [geo-blocking](https://www.joinmassive.com/glossary/what-is-geo-blocking), and no prompt engineering will get around it.

Pricing is the case with the most money attached. In a crowd-sourced study, researchers logged "the prices obtained from different vantage points and browser configurations" and found "several retailers that return prices for the same product that vary by 10%-30%," with "isolated cases that may vary up to a multiplicative factor, e.g., x2." They "could not attribute the observed price gaps to currency, shipping, or taxation differences" ([Mikians et al.](https://arxiv.org/abs/1307.4531), 2013).

The study is more than a decade old, so treat it as proof that location-based price differences exist, not as a measure of how common they are today.

## When does an AI agent need geotargeting?

Use one test: would two users in different places expect different answers to this task? If yes, the agent needs to choose a location on purpose.

| Task | Needs a location? | What to set |
|------|-------------------|-------------|
| Summarize API docs or a standards page | Usually not | Nothing |
| Answer "best X near me" for a user | Yes | The user's city |
| Track prices or stock for a market | Yes | Each market's country, and city if prices differ by region |
| Monitor brand visibility in AI answers | Yes | Each target country, plus language |
| Verify ads or offers a campaign shows | Yes | The campaign's target locations |

Datacenter IPs bring a second problem on top of the wrong location: many sites treat them differently from household connections, as our comparison of [residential and datacenter proxies for AI agents](https://www.joinmassive.com/blog/residential-vs-datacenter-proxies-for-ai-agents) explains. Our [AI chat endpoint walkthrough](https://www.joinmassive.com/blog/ai-chat-endpoint-geotargeted-llm-completions) shows how to request ChatGPT, Gemini, and Perplexity answers, with their sources, from a chosen country or city. For price work, the same rule applies to every competitor page your agent reads.

## How do you add geotargeting to an agent?

Two ways on Massive, depending on whether you want raw web access or a rendered result.

**Through the residential network.** Add geo keys to the proxy username. Per the [geotargeting docs](https://docs.joinmassive.com/residential/geotargeting), `country` is required whenever you set any geo key, and you can narrow it with `subdivision` (a state or region), `city`, or `zipcode`:

```bash
HOST=network.joinmassive.com
NAME='{PROXY_USERNAME}'
GEO=country-US-city-Chicago
CF=https://1.1.1.1

curl -x https://$HOST:65535 \
  -U "$NAME-$GEO:{API_KEY}" \
  $CF/cdn-cgi/trace
```

That target is a quick check you can run yourself: Cloudflare's trace page prints a `loc=` line with the country it sees for the exit IP, so you can confirm the location before pointing the agent at real sites.

**Through the Web Render API.** Pass `country`, `subdivision`, or `city` as query parameters on the Browsing, Search, or AI chat endpoint ([Massive Docs](https://docs.joinmassive.com/web-render/geotargeting)):

```bash
API=render.joinmassive.com

curl -G \
  https://$API/browser \
  --oauth2-bearer "$TOKEN" \
  -d url=https://example.com \
  -d country=DE \
  -d city=Berlin
```

One detail to know: on the Web Render API, if you leave `country` unset, "a random country is used by default." For any task in the "needs a location" rows above, set it explicitly.

## Verify the location you got

Don't trust the request alone. Record the location the network says it used.

- **Web Render API responses** echo `country`, `subdivision`, and `city`, the location the request was actually made from ([Massive Docs](https://docs.joinmassive.com/web-render/geotargeting)).
- **Residential HTTPS traffic** gets exit node headers in the CONNECT response, including `X-Exit-Country`, `X-Exit-Timezone`, and `X-Exit-ASN` (the exit network's autonomous system number) ([Massive Docs](https://docs.joinmassive.com/residential/exit-node-headers)).
- **Strict targeting can fail.** If no node matches your geo parameters, the residential network returns `471 No Route Match`, and per the docs, retries fail until the targeting is relaxed ([Massive Docs](https://docs.joinmassive.com/residential/error-types)). Treat that as a signal to widen from city to region, not as an outage.

Store the requested and served location with every result. When a number looks wrong later, the first question will be "where was this fetched from?" and you'll have the answer.

## Frequently Asked Questions

### Do AI agents need geotargeted web access?

Only for tasks where the answer depends on place: local search, AI answers, prices, stock, ads, and regional availability. For global reference pages, the default location is fine.

### Does ChatGPT give different answers to different people?

Yes, for two reasons. Answers can vary from one run to the next, and the location of the request shapes local answers. In our test, the same emergency-number question returned 911 in the U.S., 190, 192, and 193 in Brazil, and 119 and 110 in Japan.

### Does ChatGPT use my location?

Yes. OpenAI says ChatGPT may use an approximate location based on your IP address to provide relevant local results, and a VPN or network location may change that estimate. Sharing a precise device location is optional.

### What happens if I don't set a location on Massive's Web Render API?

A random country is used by default. Set `country`, and optionally `subdivision` or `city`, whenever the answer depends on location.

### How can I confirm which country a request came from?

Read the `country`, `subdivision`, and `city` fields in Web Render API responses, or the `X-Exit-Country` header on residential HTTPS connections.

## The bottom line

- Every request has a location. If you don't choose it, the site infers it from whatever IP the request arrives from: your server's, or an exit you didn't pick if you route through a proxy without geo keys.
- For location-dependent questions, ChatGPT gives different answers by country: our same-prompt test returned three different emergency numbers from three countries. Google, retailers, and ad systems also use location to decide what to show.
- Use geotargeting for local tasks and log the location each request got.

## Sources

- Google Search Help, ["Understand & manage your location when you search on Google"](https://support.google.com/websearch/answer/179386?hl=en)
- OpenAI Help Center, ["Searching the web with ChatGPT"](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt)
- Google, ["AI Overviews expand to over 200 countries and territories, more than 40 languages"](https://blog.google/products-and-platforms/products/search/ai-overview-expansion-may-2025-update/), May 20, 2025
- Mikians, Gyarmati, Erramilli, Laoutaris, ["Crowd-assisted Search for Price Discrimination in E-Commerce: First results"](https://arxiv.org/abs/1307.4531), 2013
- Massive Docs, [Residential geotargeting](https://docs.joinmassive.com/residential/geotargeting), [Web Render geotargeting](https://docs.joinmassive.com/web-render/geotargeting), [Exit node headers](https://docs.joinmassive.com/residential/exit-node-headers), [Error types](https://docs.joinmassive.com/residential/error-types)