How Accurate Is IP Geolocation, Really?
How Accurate Is IP Geolocation?
The honest answer: it depends entirely on what level of precision you ask for. At the country level, IP geolocation is one of the most reliable inferences in network forensics. At the city level, it is closer to a coin flip weighted in your favour. And for certain kinds of connections — mobile carriers, VPN exits, satellite links — the reported location can be confidently, systematically wrong. That is why every dossier produced by the Uplink Trace tracer carries an explicit confidence rating rather than a single authoritative-looking pin.
The numbers, without the varnish
| Precision level | Typical accuracy | Why |
|---|---|---|
| Country | 95–99% | Anchored in registry allocations, which rarely cross borders |
| Region / state | 55–80% | ISPs shuffle address blocks across regions without notice |
| City (fixed broadband) | 50–75% | Inferred from routing and latency, not from any authoritative record |
| Mobile networks | Frequently wrong | Carrier-grade NAT places traffic at a gateway city, not near the device |
| VPN / proxy | Shows the exit server | The visible IP belongs to the VPN operator's datacenter |
| Corporate networks | Often shows HQ | Branch traffic egresses through a central breakout point |
| Satellite | Can be off by entire countries | The IP surfaces at a ground station, which may be on another continent |
Why country-level is nearly always right
IP address blocks are handed out by regional internet registries — RIPE NCC, ARIN and their counterparts — to specific operators in specific countries, and that paper trail is public. An address block allocated to a German ISP overwhelmingly serves German customers, and reassignments across borders are rare and slow. Geolocation databases such as DB-IP start from this registry data and refine it, which is why the country field in a dossier deserves high trust while everything beneath it deserves scepticism.
Why region and city are much softer
No registry records which city an address serves. Region- and city-level entries are inferred: from where the operating ISP concentrates infrastructure, from latency measurements, from routing structure, and from patterns in how the provider historically deployed its blocks. Those inferences decay. An ISP can quietly move a block from Lyon to Marseille during a network reorganisation, and the databases will keep reporting Lyon until fresh evidence accumulates. On fixed broadband the city is right roughly half to three-quarters of the time — useful for triage, never sufficient for a conclusion. The mechanics are covered in more depth in how IP geolocation works.
Mobile networks and carrier-grade NAT
Mobile carriers do not have enough IPv4 addresses to give every handset its own. Instead they use carrier-grade NAT (the shared address space defined in RFC 6598), funnelling thousands of subscribers through a small pool of public addresses at a regional gateway. The IP you observe belongs to that gateway, which may sit hundreds of kilometres from any individual subscriber. A phone in a small town routinely geolocates to the carrier's gateway in the nearest major city. When a trace lands on a mobile carrier's AS, treat the city field as the carrier's plumbing, not the user's whereabouts.
VPNs, proxies and datacenter ranges
A VPN does exactly what it promises: the visible IP is the exit server's, so geolocation reports the exit server's datacenter — accurately. The database is not failing; it is correctly locating a machine that is not the one you care about. Uplink Trace flags datacenter and VPN ranges heuristically from the AS operator's name, and labels them LIKELY rather than asserting proof, because an operator name is evidence, not certainty. Understanding what an ASN is helps here: the autonomous system tells you who operates the network, which is often more informative than the coordinates.
Corporate networks and satellite links
Large organisations frequently route branch-office traffic through a central internet breakout, so an employee in Calgary can surface with an IP registered to headquarters in Toronto. Satellite internet is the extreme case: traffic reaches the terrestrial internet at a ground station, which may be in a different country — occasionally a different continent — from the dish. Both patterns produce geolocation results that are internally consistent and operationally misleading.
Why accuracy varies by country
Accuracy is not uniform across the map. Small countries are trivially accurate at the country level and often at the city level, because there is simply less territory to be wrong about. Countries with dense, competitive ISP markets and regionally structured allocations — much of Western Europe, urban North America — geolocate comparatively well. Countries where a national carrier draws from one countrywide address pool geolocate poorly below the country line, because nothing in the addressing structure distinguishes one city from another. The same database can be excellent in the Netherlands and nearly useless at city level elsewhere.
How to read a confidence radius
An honest geolocation result is a circle, not a point. A pin on a map asserts precision the underlying data cannot support; a circle says the network evidence places this address somewhere in this probable area. Read the radius as an error bar: a tight circle over a city on a fixed-broadband ISP is a reasonably strong signal; a wide circle centred on a mobile gateway or a datacenter is the tool telling you the centre point is close to arbitrary. When Uplink Trace draws a broad radius, that is the finding — narrowing it further would be invention. You can see how this renders on a well-documented address in the example trace of 8.8.8.8.
What this means in practice
Country-level results are dependable enough for compliance screening, log triage and abuse analysis. Region and city results are leads to corroborate, not facts to cite. And none of it identifies a person: IP geolocation approximates where a network attaches to the internet, not who is using it or where they live — a distinction unpacked in can someone find my address from my IP. Any tool that shows you a precise pin without a confidence rating is not more accurate than the data allows; it is merely more confident than it should be.