How Accurate Is IP Geolocation? Real Numbers by Country, Provider, and Connection Type

Finland: 38%. The United States: 63%. Kuwait: 100%. These are city-level IP geolocation accuracy rates from the same provider, using the same database, queried on the same day. The "how accurate is IP geolocation" question doesn't have one answer — it has hundreds, and most of them aren't what people expect.

Marketing pages quote "99% accuracy" because that's true at the country level. But country-level accuracy is the easy part. The moment you need to know which city someone is in, accuracy drops to a range so wide it becomes nearly meaningless without context. A lookup in Singapore is right 99% of the time. The same lookup in Hong Kong is right 25% of the time. Same technology, same database, wildly different results.

If you need the basics of how IP geolocation works — the six data sources, the database architecture, the allocation pipeline — we covered that in detail in our IP geolocation explainer. This article skips the "how it works" and goes straight to the numbers: what's accurate, what isn't, and why it varies so dramatically depending on where you are, how you connect, and which database you're using.

The Three Tiers of Accuracy

IP geolocation accuracy isn't a single number. It depends entirely on what level of geographic precision you're asking about. There are three distinct tiers, and the drop-off between them is steep.

• Country level (95–99.8%) — Almost always correct. MaxMind reports 99.8% for their GeoIP2 database. Even free databases reliably identify which country an IP belongs to. This is the number that "99% accuracy" marketing claims reference, and it's legitimate.
• Region/State level (55–80%) — Varies dramatically by country. MaxMind reports ~80% for US states, but IEEE research found rates as low as 20% in some countries. The problem compounds in countries where ISPs centralize traffic routing through a single city.
• City level within 50 km (38–86%) — This is where accuracy claims start to diverge from reality. According to MaxMind's own published accuracy data, city-level accuracy in the US is 63%. In Finland, it drops to 38%. In Ethiopia, it's 86%. The variation is enormous and depends on geography, infrastructure density, and how ISPs allocate their address space.

The practical implication: if someone tells you IP geolocation is "99% accurate," ask "at what level?" Country-level accuracy is genuinely excellent. City-level accuracy is a coin flip in some countries.

Accuracy by Country: The Full Data

This is the data that most articles about IP geolocation accuracy don't include. MaxMind publishes a city-level accuracy comparison tool for their GeoIP2 City database, broken down by country. These are their own reported numbers — not our testing, not a third party's estimate. The variation across countries is striking.

The pattern is not random. There's a clear structural explanation for why some countries score so much higher than others.

Why small countries win

Kuwait, Singapore, Bahrain, Qatar, and Malta share a common trait: they're geographically tiny. Kuwait is 17,818 km² — smaller than New Jersey. Even if the geolocation database maps an IP to the wrong part of the country, it's still within 50 km of the right answer because the entire country fits inside that radius. A "miss" in Singapore might be off by 15 km. A "miss" in the United States could be off by 3,000 km.

Why Scandinavia struggles

Finland and Sweden both score 38%, despite being wealthy, technologically advanced countries. The reason is infrastructure centralization. Nordic ISPs route traffic through a small number of major hubs (Helsinki, Stockholm), but subscribers are spread across vast, sparsely populated areas. An IP that exits through Helsinki might belong to a subscriber 500 km north in Oulu. The geolocation database correctly identifies the ISP's routing point, but that routing point is nowhere near the subscriber.

Why the US isn't better

At 63%, the United States performs below what most people assume. American ISPs are large and geographically distributed, but they also engage in frequent IP block transfers, CGNAT deployments, and regional consolidation that make it harder for geolocation databases to keep mappings current. The University of Chicago study found median errors of 2.0–4.0 km in major metropolitan areas like New York and Chicago — but that's specifically for fixed-line broadband in dense urban areas. Rural and cellular accuracy is much worse.

Connection Type Matters More Than Provider

Most discussions about IP geolocation accuracy focus on which database provider is "best." The data suggests a different variable matters more: how the user connects to the internet.

Broadband / WiFi

Fixed-line broadband connections produce the most accurate geolocation results. ISPs assign IP blocks to specific regions, and the subscriber's traffic exits through local or regional infrastructure. The University of Chicago study measured median errors of 2.0–4.0 km for broadband connections in major US cities — genuinely useful precision for most applications. Outside of dense urban areas, accuracy degrades, but broadband remains the most reliably geolocated connection type.

Cellular / Mobile

Cellular networks are fundamentally different. Mobile carriers route traffic through regional switching facilities that can be hundreds of kilometers from the subscriber. A T-Mobile user in rural Montana might have their traffic exit through a gateway in Denver. The geolocation database correctly identifies the gateway's location, but that's not where the person is.

CGNAT compounds the problem. Mobile carriers aggressively use Carrier-Grade NAT, meaning hundreds or thousands of subscribers share a single public IP address. That shared IP maps to one location — the carrier's NAT gateway — regardless of where individual subscribers actually are. According to research from IEEE, 33–80% of mobile IP samples were assigned to the wrong administrative region, depending on the country.

Satellite (Starlink)

Starlink presents the worst case for IP geolocation. With over 4 million subscribers globally, Starlink routes traffic through ground stations that can serve areas spanning hundreds of miles. The IP address maps to the ground station, not to the subscriber's dish on their roof. A Starlink user in rural Wyoming might geolocate to a ground station in Oregon. Inter-satellite laser links can route traffic to exit the constellation from a ground station on a different continent entirely.

Starlink officially acknowledges this problem. APNIC has documented cases where Starlink IPs geolocate to entirely the wrong country. As satellite internet adoption grows, this failure mode affects an increasingly significant share of internet traffic.

IPv4 vs IPv6: The Growing Accuracy Gap

IPv4 addresses have been assigned, tracked, and mapped for decades. Geolocation databases have had 30+ years to build comprehensive, battle-tested mappings. IPv6 is a different story.

Research from Radboud University found that IPv4 geolocation achieves approximately 90% country-level accuracy, while IPv6 drops to 40–60%. APNIC's analysis found 77% of IPv4 economy-level mappings were accurate versus 65% for IPv6. The gap exists because IPv6 address blocks are newer, assigned in much larger prefixes, and have less historical data for databases to work with.

This matters because IPv6 adoption has passed 40% of global traffic, according to Google's IPv6 statistics. A growing share of internet users are on a protocol where geolocation is measurably less accurate — and most users don't know which protocol they're on at any given moment.

The gap is closing as databases accumulate more IPv6 mapping data, but "closing slowly" still means millions of inaccurate lookups per day right now.

Provider vs Provider: Who Is Most Accurate?

Several independent studies have compared major geolocation providers head-to-head. The results are less decisive than you might hope.

ipapi.is comparative study (January 2026)

This study tested 10 geolocation API providers against GPS ground truth from over 2,000 residential IPs. Country-level accuracy averaged ~92% across providers. City-level accuracy (within 50 km) ranged from 50% to 75%. At neighborhood level (within 10 km), accuracy dropped to 15–35%. The study found meaningful differences between providers at the city level, but no single provider dominated across all regions.

IEEE BalkanCom 2023

This academic study compared MaxMind, DB-IP, IP2Location, and IPGeolocationIO across 6.3 million unique IPs. The most striking finding: the average pairwise discrepancy between databases was 620 km. That means for any given IP, two different reputable databases would disagree on the city by an average of 620 kilometers. The databases agreed on country most of the time but diverged sharply at finer granularity.

University of Chicago: paid vs. free

This GPS-validated study produced a finding that surprised many in the industry: paid database versions were not significantly more accurate than free versions for residential broadband IPs. The median error difference between MaxMind's paid GeoIP2 City and their free GeoLite2 was small in the US cities tested. The primary advantages of paid tiers turned out to be update frequency and additional data fields (ISP, connection type, proxy flags), not fundamentally better coordinate data.

Database Freshness: The Degradation Problem

IP geolocation databases are snapshots. The internet is not static. Between updates, the mapping between IP addresses and locations degrades as ISPs reassign blocks, networks reconfigure, and address transfers occur.

• IP2Location reports 1–5% accuracy loss per month without updates. They estimate 5–10% of their records change monthly.
• MaxMind experiences approximately 1.5% degradation per month, according to their documentation.
• Practical impact: A database that's 6 months old could be 10–30% less accurate at the city level than a current one.

Update frequencies vary significantly between providers:

This has a practical implication that most users miss: when you check an IP lookup tool and the result seems wrong, the issue might not be the database provider's methodology. It might simply be that the IP was reassigned since the last database update. Running a traceroute alongside a geolocation lookup can sometimes reveal the discrepancy — the traceroute shows the actual network path while the database shows the historical mapping.

Is This Accurate Enough? Decision Framework by Use Case

Accuracy requirements vary wildly by application. A 63% city-level accuracy rate is useless for precise targeting but perfectly adequate for content licensing. The question isn't "is IP geolocation accurate?" but "is it accurate enough for what I'm trying to do?"

The pattern is clear: country-level applications work. City-level applications work with caveats. Anything requiring precision beyond the city level requires a different technology entirely — GPS, Wi-Fi triangulation, or ISP records with legal process.

What's Improving and What's Getting Worse

IP geolocation accuracy isn't on a simple upward trajectory. Some factors are pushing accuracy higher while others are actively degrading it. The net result depends on which trends dominate for a given user population.

Getting better

• RFC 8805 geofeeds: ISPs can now self-publish CSV files mapping their IP prefixes to locations. Google already incorporates geofeeds into their pipeline. As adoption grows, the most common source of error — ISPs whose IP assignments don't match registry data — will gradually resolve. More on this in our IP geolocation overview.
• GPS ground truth from mobile: Providers like Digital Element collect GPS coordinates from 1.5 billion mobile devices per 30-day window, generating 60–70 million data points daily. This creates continuously validated IP-to-location mappings that improve the databases for everyone.
• Faster updates: MaxMind updates their paid database every weekday. DB-IP updates tens of thousands of records daily. The gap between reality and the database is shrinking, at least for premium tier customers.
• Machine learning classification: Providers increasingly use ML models to cross-reference multiple data sources (BGP, DNS, latency, telemetry) and improve confidence scoring, particularly for ambiguous IPs where traditional methods produce conflicting signals.

Getting worse

• CGNAT growth: Over 100 million users worldwide share IP addresses via Carrier-Grade NAT. As IPv4 exhaustion continues, more ISPs deploy CGNAT, making shared IPs even harder to geolocate accurately. A single IP might represent users spread across an entire metropolitan area.
• VPN adoption: An estimated 1.75 billion people use VPNs globally, according to APNIC research from December 2025. VPN traffic geolocates to the exit server, not the user. This doesn't make geolocation databases less accurate per se, but it means a growing share of traffic produces deliberately misleading results.
• Starlink and satellite: With 4 million+ subscribers and aggressive expansion plans, satellite internet creates a growing population whose IPs map to ground stations instead of user locations. SpaceX's ambitions suggest this will be tens of millions of subscribers within a few years.
• IPv6 transition: As IPv6 carries more traffic, the less mature geolocation data for IPv6 addresses means aggregate accuracy across all traffic is diluted. The 40–60% country accuracy for IPv6 (vs 90%+ for IPv4) pulls the blended average down.
• IP address market transfers: IPv4 addresses are now routinely bought and sold between organizations, sometimes across continents. An IP block registered to a company in Virginia might be transferred to one in Singapore. Until all databases update, that block geolocates incorrectly.

Test It Yourself

The best way to evaluate IP geolocation accuracy is to test it against your own known location. Here's how.

• Check your own IP: Use our free IP lookup tool to see what your IP resolves to. Compare the city, region, and country against where you actually are. If you're on broadband, it should be close. If you're on cellular or satellite, it might be off.
• Check VPN impact: Connect to a VPN and re-run the lookup. The result should show the VPN server's location, not yours. Then run it through our VPN detector to see if the VPN IP is detected as a VPN. This demonstrates both the geolocation result and the VPN detection layer.
• Compare providers: Look up the same IP on multiple geolocation services. If they all agree, confidence is high. If they disagree on the city, that's the 620 km discrepancy problem in action.
• Test it on real traffic: If you log visitors with a tracking link, geolocate a handful of the captured IPs and compare the cities against what those visitors actually report — the accuracy spread becomes obvious fast.
• For developers: Our geolocation API returns structured data including country, city, coordinates, ISP, connection type, and ASN. Test programmatically against known-location IPs to evaluate accuracy for your specific use case.

For a deeper understanding of what IP geolocation can and can't reveal about any given IP address, see our guide on tracing an IP address and the taxonomy of IP tracking tools.

Frequently Asked Questions

Sources: Accuracy data cited in this article is sourced from MaxMind GeoIP2 City Accuracy Comparison, University of Chicago GPS-Based Geolocation Study, IEEE BalkanCom 2023 (6.3M IP comparison), IEEE Transactions on Mobile Computing (Spain/France/UK study), ipapi.is Independent Comparative Study (2026), Radboud University IPv4/IPv6 Accuracy Study, IP2Location Data Accuracy, RFC 8805 (Geofeeds), and APNIC Starlink Geolocation Analysis.