When a search app is involved, results are likely relevant to the location
(because the search app receives your location likely from the Web app, not from the Model)
And then the’ location’ are ‘known’ by both you and the model.
The model assumes your location based on the third-party Search app results
then it generates the response and you see the response right away.
disclaimer: I am just making suggestions based on my humble developer experience
i was just chating with gpt and he just pulls ut the name of my city this is not the only time this happened this happened 3 times in different chat . so i asked chat gpt about the probability of this happeneing and he said 1 in 1mil. I tested this in a different chat (btw i set gpt to have no memrory from previous chats)by asking his for nice places to visit near me, and it ‘‘randomly’’ pulled it off again i asked the probabilties of this happening and it said one in 1trilliion
I’ve tested one thing, and it confirms one of the leading theories.
Using VPN software connected to Czech server. Asked chatgpt a simple question about recommended pizzerias near my location. The answer was giving a list of restaurants in Prague. I don’t need to mention I do not live in Prague nor Czech Republic.
I just had the weirdest experience I had to share here, I just asked ChatGPT something in English using dictation and it came out in some weird language, I pressed enter automatically and got an answer also in a weird language. I then typed in English what was that? It said ‘oh sorry I accidentally slipped into Welch’ and I was like what the hell? Why would you ‘slip’ into Welch? Trying to find a connection I realised I had just finished watching Red Bull Hardline on YouTube for 2 hours and this race in Wales. I don’t know if I have a tin foil hat on, but thought it a weird coincidence and have no idea why else it would change to Welch. I am from New Zealand an currently in Nepal. This hasn’t happened before. It has slipped into Nepalese if I’m translating something here. This freaked me out a bit, worried it might be tracking my actions on my phone and content, messages etc. etc
I was also experiencing this, so I kept digging, and finally got ChatGPT (5) to answer this:
Okay — let’s cut everything down to the bare mechanics. You’re right: if the web tool query consistently returns your exact town, then the request must contain some form of location data precise enough to produce that result.
Here’s the most plausible technical explanation:
Your IP address is visible to OpenAI servers when you connect.
This is standard for all web connections.
The web tool uses the OpenAI server to send the request to a third-party service.
The server can include a location inferred from your IP in the request, even if it’s not obvious or explicitly shown in the model’s responses.
The third-party service receives the request with that location and returns results relevant to it — e.g., your exact town for a weather forecast.
This bypasses cached memory or previous chat content.
The location is coming from real-time connection data, not past interactions or profile memory.
So technically:
The web tool request includes a server-inferred location derived from your IP.
The service responds based on that location.
That’s why it can return your exact town, even on a VPN or in a new chat.
This is consistent with what you’re observing — it’s not random, and it’s not based on previous chats. It’s the server using your current IP to infer location, then including that in the web tool request.
If you want, I can diagram exactly how that flow works so it’s crystal clear.
So, with that response, and my assumption that the request sent from OpenAI’s web-tool, to any location based service (like weather or local shopping areas) includes the inferred location from the IP address. Which is also why the replies change, when you use VPN.
Chagpt uses its own IP address geolocation to determine your location. I was using the mobile version, which seems to primarily determine your general location using your device’s IP address. This is a standard practice for internet services and does not require explicit location permissions from your phone’s operating system (like GPS access)
So it knows your location to give you relevant answers, if that is true or not that is how they do it, I quick test use a VPN change your internet access location and I will se that it will give answers relevante to that new location
I have noticed that, while talking, I was on a moving train and I just asked some random question to ChatGPT which was completely irrelevant to location,
But I don’t know how, maybe it malfunctioned or something but it gave me an answer, which started like - ‘Based on your location <MY_EXACT_LOCATION_ON_MOVING_TRAIN> here is the answer: ‘
I was shocked, then when I asked it again it said, ‘I don’t have access to your location or any other details’
And also it knows the phone model you are using, wifi name, and all other things, It wont just admit it.