DEV Community

Cover image for IP geolocation accuracy: why databases disagree on your IP
proxymint
proxymint

Posted on Originally published at proxymint.com

IP geolocation accuracy: why databases disagree on your IP

Originally published on proxymint.

IP geolocation accuracy by level, why two databases put one IP in different cities, how geofeeds work, and how to check a proxy exit's location yourself.

TL;DR

  • Trust the country, be careful with the city. MaxMind puts its own country accuracy at 99.8% but only 66% for US cities within 50 km (checked October 2026), so a city label that is off by one suburb is normal, not a bug.
  • Two lookup sites, two cities is the usual case. Each database mixes registry records, operator geofeeds, latency probes and its own data, and stores ranges of different sizes, so they split at city level all the time.
  • Mobile is the weak spot. A 2026 study of four providers found median mobile errors of 179-207 km, so an empty city on a carrier address is expected, and region is the level worth reading.
  • The target's database is the one that counts. If the exit is right and the page is still wrong, check DNS first: plain socks5:// resolves on your machine, which is the annoying one to spot.

Key takeaways

  • Read accuracy_radius, not just city.name. A city with a 500 km radius is a region guess, and code that ignores the radius throws the warning away.
  • Run the two-database check on your own exits. Look up 200 exits in an echo service and a local GeoLite2 file, and count country, region and city agreement before you scale a geo job.
  • Fix DNS before you blame the database. Use an http:// or socks5h:// proxy URL so the target name resolves on the proxy side, then set browser language and time zone for the market.
  • Target the level the page varies by. National prices need only country; local search and store pages need city, checked against the exact page you are collecting.

Read the full guide on proxymint, with the commands, tables and sources.

Top comments (0)