How IP Geolocation Works: Inside GeoIP Databases

Look up an IP on IPchase and within milliseconds you get "China · Fujian · Ningde · China Unicom." Many people wonder: where does this data come from? How accurate is it? Why does it sometimes point to the neighboring province?

The Core Logic: A Map of IP Ranges

IP addresses are not registered device by device — they are allocated in blocks. IANA (the Internet Assigned Numbers Authority) hands address blocks to the regional registries (APNIC covers the Asia-Pacific), and registries allocate further to carriers and companies.

Every allocation record says: who this range was assigned to, and which city it was registered in. Aggregate the allocation data, the routing announcements, and real-world measurements, and you get a GeoIP database.

Under the hood the database is a prefix tree: a query for 1.2.4.8 walks down the binary prefixes, hits a node, and the node says "this range belongs to Ningde, Fujian."

The Major Data Sources

MaxMind GeoLite2: the de facto standard. The community edition is free, city-level accuracy, updated weekly. It's what IPchase uses, deployed locally on our servers
IP2Location / DB-IP: commercial databases, each with its own data corrections
CZ88 (Chunzhen) IP database: a long-established Chinese database with solid supplementary coverage of mainland China

Accuracy: City Level Is the Ceiling, Not a Guarantee

Set expectations first: GeoIP's city-level accuracy runs roughly 70%–90%, and that's the ceiling.

The reason: IP registration records the carrier's registered address, not the user's physical location:

An IP range assigned to a county town may be registered under the prefecture-level city branch
Mobile traffic exits through provincial gateways, so users across the whole province look up as the provincial capital
Anycast IPs on CDNs (like 8.8.8.8) are shared by dozens of nodes worldwide; the honest answer is "anycast"

So discrepancies like "it says Ningde but I'm in Fuzhou" are inherent to how the data works. Every GeoIP service has them — there is no such thing as an absolutely accurate database.

The Update Cycle: Why Weekly Refreshes Matter

Carriers keep reshuffling their ranges. MaxMind publishes a new GeoLite2 build every Tuesday, which includes:

Newly allocated ranges
Corrected ownership records
ASN (autonomous system) changes

Let the database go stale and the consequences show: new ranges return blank, or ownership still points to the previous holder.

IPchase runs a local deployment on our own servers with daily automatic update checks — lookups depend on no external API, respond fast, have no query limits, and the database is pulled automatically whenever a new version ships.

What Else a Lookup Returns

A single IP lookup carries more metadata than most people expect:

Field	Source	Purpose
Country/province/city	GeoIP City database	Displaying geolocation
Latitude/longitude	City database (city center point)	Map markers
Time zone	City database	Showing the local clock
ASN / carrier	GeoIP ASN database	Telling datacenter from residential
IP type	Datacenter/residential flags	Key to proxy detection

ASN deserves special mention: AS4134 is the China Telecom backbone, AS15169 is Google — the ASN tells you whose network an IP belongs to. It's one of the core inputs for proxy detection.

Typical Cases of "Off by a Province" Results

Once you know the mechanism, the usual discrepancies all fall into place:

Mobile traffic: exits are consolidated at provincial core gateways, so users across the province present the same city — the single largest source of "looked up as the provincial capital" results
Newly provisioned broadband: registration lags behind newly activated ranges, so the GeoIP database may still show the previous holder's location
Corporate lines: headquarters in Shanghai, branches using ranges registered to HQ — everyone looks up as Shanghai
VPN / proxy: you're seeing the exit server's location, not yours — that's how proxies are supposed to work
Satellite networks: for services like Starlink, the gateway station can be over a thousand kilometers from the user

To tell which case you've hit, look at the ASN: if it belongs to a mobile carrier (China Mobile, Vodafone, and the like), it's most likely gateway consolidation; if it belongs to a cloud or VPS provider (Alibaba Cloud, DigitalOcean), you're seeing the datacenter's location.

The Coordinates Are the City Center, Not Your House

The latitude and longitude a GeoIP City database returns are the coordinates of the registered city's center — the precision is the city itself. A point on the map two kilometers from your home is normal; two provinces off is also normal (a registration discrepancy).

Some commercial databases offer street-level positioning, but that comes from layering in WiFi positioning data, user reports, and other signals — no longer "pure IP geolocation." The ceiling for pure IP lookup is the city. Keep that in mind and you won't over-expect from any lookup service.

How to Cross-Check

To raise your confidence in a geolocation judgment, look at three kinds of signals together:

GeoIP location — what the registration data says
Time zone and language — whether the browser's time zone and system language match the geolocation (that's exactly what the disguise score test checks)
Latency corroboration — use the Ping tool to measure RTT to the claimed city and see whether it's plausible for the physical distance: Singapore to Los Angeles has a physical floor of roughly 160ms, so a 40ms reading means the "Los Angeles" attribution is almost certainly wrong

Three signals corroborating each other beat any single database.

Summary

IP geolocation = allocation records plus routing data; city-level accuracy of 70–90%
Lookups are prefix matching against a local database, done in milliseconds
The data changes weekly; when choosing a service, look at its update mechanism
The ASN is the "clan" behind an IP — more informative than the city attribution

Try it: the IP lookup tool — enter any IP and see its full profile.