lpellegr
Joined 11 years ago
Joined 11 years ago
ChatGPT receives your IP when you connect. The model itself does not know your location, but the system around it can add location context before the model answers. Many products do this because it helps with weather, time zone handling or local recommendations.
Accuracy depends on your network setup. VPNs, mobile networks and corporate proxies can make the location drift quite a bit, but at a general level it is often good enough to shape responses.
The more important part is transparency. Inferring location from an IP is normal, but it should be clear to users when that information is being used.
A great IP intelligence solution is also ipregistry.co
Running an opensource AI model to perform translation may help and work in some cases but to be honest this has also a lot of limitations. The first being processing time and the second languages support. I am not even considering accuracy.
Measurement-based probing has been used by major IP intelligence providers for years. Some just market it better than others But it’s not a silver bullet, a lot of IPs (mobile, CGNAT, enterprise) aren’t consistently probeable or yield ambiguous results.
In practice, no single method wins. Hybrid approaches usually perform better.
(For transparency: I work for https://ipregistry.co, worth including in your benchmark if you’re comparing providers.)