MaxMind, the company behind one of the most widely used IP geolocation databases in the world, publishes a number that most people in the industry gloss over: their GeoIP2 product achieves 99.8% accuracy at the country level, but only about 66% at the city level for U.S. IP addresses within a 50-kilometer radius. That is a 34-percentage-point drop just by changing the question from "what country?" to "what city?" — and it gets worse from there depending on where in the world you are looking.

This gap between country-level and city-level accuracy is the single most misunderstood aspect of IP geolocation. Marketing pages from geolocation providers prominently display the 99%+ country number. Blog posts recycle it without context. The result is that developers, fraud analysts, and compliance teams build systems with assumptions about precision that the underlying data does not support.

This article breaks down what the accuracy numbers actually mean, where the data comes from, and what you can realistically expect from an IP geolocation lookup depending on the resolution you need, the region you are querying, and the type of IP address you are looking at. No inflated claims — just the published benchmarks from the major providers and the peer-reviewed research that tests them.

What "Accuracy" Actually Means in IP Geolocation

When someone asks "how accurate is IP geolocation?" the answer depends entirely on what level of geographic resolution they are asking about. There are four distinct levels, and accuracy drops dramatically as you move from coarse to fine.

Country level (99-99.8% accurate)

This is the resolution where IP geolocation genuinely works well. IP address blocks are allocated by Regional Internet Registries (RIRs) — ARIN for North America, RIPE NCC for Europe and the Middle East, APNIC for Asia-Pacific, AFRINIC for Africa, and LACNIC for Latin America. Each allocation is tied to a country. Since these registrations are public and well-maintained, determining what country an IP belongs to is a solved problem for the vast majority of addresses.

The remaining 0.2-1% error margin comes from edge cases: IP blocks that have been transferred between organizations in different countries, satellite internet users whose IP is registered to the ground station rather than their physical location, and VPN or proxy servers that route traffic through a different country.

Region/state level (70-85% accurate)

At the state or province level, accuracy begins to diverge. MaxMind estimates approximately 80% accuracy at the state level for U.S. IPs. The challenge here is that ISPs do not always register their IP allocations with state-level granularity, and some ISPs serve customers across multiple states from centralized infrastructure.

City level (50-80% accurate, depending on region)

This is where the numbers get uncomfortable. Even the best commercial databases achieve city-level accuracy of only 66% for U.S. IPs within a 50-kilometer radius, according to MaxMind's published testing. IP2Location reports over 75% city-level accuracy globally. DB-IP claims over 97% at the city level, though their measurement methodology and radius threshold differ.

The discrepancies between providers are not just marketing — they reflect genuinely different testing methodologies, different ground-truth datasets, and different definitions of "accurate." When one provider measures accuracy within a 25-kilometer radius and another uses 50 kilometers, the numbers are not directly comparable.

Postal code and coordinate level (unreliable)

Postal code accuracy is generally below 50%, and specific latitude/longitude coordinates from an IP lookup should be understood as a general area indicator, not a pinpoint. The coordinates typically resolve to a city center, an ISP facility, or a default location for that region. As we covered in our guide to IP location maps, the pin on a map represents the database's best estimate for the IP block — not where any specific person is sitting.

Real Accuracy Benchmarks by Provider

The three major commercial IP geolocation database providers — MaxMind, DB-IP, and IP2Location — each publish accuracy data, but they use different testing methodologies, which makes direct comparison tricky. Here is what each provider publishes, along with what independent research says.

Why the numbers differ so much

DB-IP's claimed 97%+ city-level accuracy versus MaxMind's 66% is not necessarily a contradiction. The differences come down to three factors:

• Radius threshold. What counts as "correct"? If you define city-level accuracy as "within 100 kilometers," you will get a much higher number than "within 25 kilometers." MaxMind uses 50 kilometers. Others use larger thresholds or do not specify.
• Ground-truth dataset. Accuracy testing requires known IP-to-location pairs. The quality, geographic distribution, and size of this ground-truth data directly affects the result. A test set skewed toward urban, well-connected areas in developed countries will produce higher accuracy numbers than one that includes rural IPs from developing regions.
• IP type composition. A test set with mostly residential broadband IPs will score higher than one that includes mobile, VPN, and data center IPs. Mobile IPs alone can drag city-level accuracy down to 40-65%.

What independent research says

Academic studies provide a more sobering picture. A deep-dive analysis published on arXiv that tested two commercial geolocation databases against a ground-truth dataset of over 2 billion samples found that while country-level geolocation was correct in most cases, at finer granularities — state, city, and postal code — at least 33%, 40%, 70%, and 80% of samples respectively failed to resolve to the correct administrative region. A 2023 study published in MDPI's Electronics journal found that a fusion approach combining multiple databases could achieve 94% city-level accuracy with 99.99% city coverage, but that was using combined sources rather than any single database.

Country level  → Trust it. Build on it. (99%+)
Region/state   → Mostly reliable, verify for edge cases. (70-85%)
City level     → Useful directional signal, not ground truth. (50-80%)
Postal code    → Do not make business decisions on this alone. (<50%)
Coordinates    → General area only. Never treat as a precise address.

Accuracy Breakdown by Region

Geography is the biggest variable in IP geolocation accuracy after resolution level. The difference between looking up an IP in Germany versus one in Nigeria is enormous, and it comes down to infrastructure density and ISP registration practices.

Why the gap exists

The accuracy disparity traces to three structural factors:

ISP infrastructure density. In the United States or Germany, ISPs serve relatively small geographic areas from local facilities. A Comcast node in Denver almost certainly serves Denver metro users. In contrast, an ISP in Kenya might serve a 500-kilometer radius from a single point of presence in Nairobi. The database maps the IP to Nairobi, but the actual user could be in Mombasa.

Registration practices. RIPE NCC (which covers Europe) requires detailed geographic annotations when ISPs register IP allocations. AFRINIC's registration data tends to be coarser, often listing only the country and capital city. This cascades into the geolocation databases that rely on this registration data as a foundation.

Mobile vs. fixed-line ratios. In many developing regions, a much higher percentage of internet access is mobile-only. Mobile carriers use CGNAT pools that serve wide areas, which inherently reduces geolocation accuracy. In Sub-Saharan Africa, mobile internet accounts for a significant majority of connections, while in Western Europe and Japan, fixed broadband remains common.

Factors That Affect Accuracy

Beyond geography and resolution level, several technical factors determine whether a specific IP lookup will be accurate or misleading.

VPNs and proxy services

Approximately 22.9% of internet users worldwide used a VPN in 2024, according to data aggregated by multiple research firms. In the United States, that figure is higher — roughly 42% of Americans report having used a VPN. When a user connects through a VPN, the IP address seen by the destination server belongs to the VPN provider's server, not the user. The geolocation lookup correctly identifies where that server is, but it tells you nothing about where the user physically sits.

This is not a database error. It is an inherent limitation: IP geolocation maps network paths, not human beings. For fraud detection purposes, a lookup showing a VPN IP is actually useful information — it tells you the user is obscuring their location, which is itself a risk signal that tools like InfoSniper's IP reputation checker can flag.

Carrier-grade NAT (CGNAT)

IPv4 address exhaustion has forced ISPs worldwide to deploy CGNAT, which allows hundreds or thousands of customers to share a single public IP address. The geolocation database maps that IP to wherever the CGNAT device is located — typically at a regional point of presence. A user in a suburb 80 kilometers away will show the same IP and the same mapped location as someone sitting next to the CGNAT equipment.

CGNAT deployments have grown steadily. Research tracking CGNAT deployment found the number of observed deployments grew from 1,200 in 2014 to over 3,400 in 2016, with nearly 29% of those in mobile operator networks. The growth has continued since as IPv4 addresses become more scarce and expensive — trading at $35 to $60 per address as of early 2025.

Mobile vs. fixed-line connections

Mobile IPs are consistently less accurate than fixed-line residential IPs. Mobile carriers route traffic through regional hubs, and the IP address pool associated with a particular cell tower often covers a metropolitan area or larger. An IP lookup for a T-Mobile user in a Phoenix suburb might show "Phoenix" (close enough) or "Denver" (the regional routing hub — not close at all).

The difference is significant: city-level accuracy for fixed-line residential IPs in developed countries typically ranges from 70-85%, while mobile IPs in the same countries drop to 40-65%.

Corporate and cloud IP addresses

Large organizations route all employee traffic through centralized data centers. A company with headquarters in San Francisco and offices in five other cities might have all traffic exit through a San Francisco IP. Cloud-hosted applications and data center IPs present a different challenge: the IP maps to the data center's physical location, which may have no relationship to the end user.

If you look up an IP and it resolves to AWS us-east-1 in Virginia, that tells you where the server is, not where the person using it is. The ISP and AS number fields from a tool like InfoSniper help distinguish these cases — seeing "Amazon Technologies" as the ISP immediately tells you this is cloud infrastructure, not a residential connection.

Satellite internet

Services like Starlink present a unique challenge. The IP address is assigned to a ground station, which can be hundreds of kilometers from the user's actual location. A Starlink user in rural Montana might have their IP mapped to a ground station in Washington state. This is a growing edge case as satellite internet adoption increases in rural areas where traditional ISP infrastructure is sparse.

IPv4 vs. IPv6 Accuracy Differences

IPv6 geolocation is measurably less accurate than IPv4, and this gap matters as IPv6 adoption continues to grow.

A research study comparing IPv4 and IPv6 geolocation accuracy found that IPv4 addresses achieved approximately 90% country-level accuracy while IPv6 addresses scored only 40-60% in some databases. The study also found that constraint-based geolocation (CBG) showed up to 30% larger average error distances for IPv6 compared to IPv4, with performance varying by region — European hosts performed best, while Asia-Pacific hosts performed worst.

Three factors drive the gap:

• Database maturity. IPv4 geolocation has been refined over more than two decades. IPv6 databases are significantly younger. There are simply fewer verified data points for providers to train their models on.
• Allocation block size. ISPs often receive enormous IPv6 blocks (a /32 gives an ISP 79 billion /64 subnets) and may not annotate geographic assignments at a granular level. A single IPv6 allocation might cover an entire country without sub-allocations that indicate city-level deployment.
• Adoption patterns. IPv6 adoption is uneven. Countries with high adoption — India, the US, Germany — tend to have better IPv6 geolocation data. Countries where IPv6 is rare have minimal data to build accurate mappings.

How Geolocation Databases Maintain and Improve Accuracy

Understanding how the databases are built explains why accuracy varies and what the providers are doing to improve it.

Data sources

Every major geolocation provider builds on a foundation of several data sources:

• RIR registration data. ARIN, RIPE NCC, APNIC, AFRINIC, and LACNIC maintain public records of IP block allocations. This is the baseline for country-level mapping and provides the initial geographic signal.
• ISP data feeds. Some ISPs share information about which IP ranges serve which cities. The quality and detail of this information varies enormously by ISP and by region.
• BGP routing analysis. By analyzing Border Gateway Protocol routing tables and traceroute paths, providers can infer where network infrastructure is physically located based on how traffic flows.
• Active network probing. Latency measurements from distributed probes can triangulate the approximate location of an IP endpoint. If probes in New York, Chicago, and Atlanta all measure specific latency to an IP, the distances constrain where that IP can physically be.
• GeoFeeds. RFC 8805 introduced a standardized way for network operators to publish geographic information about their IP ranges. A 2024 study found that while GeoFeeds have the potential to improve accuracy, adoption is still limited and the data quality varies.
• User-confirmed locations. Some providers incorporate anonymized, user-consented location data from mobile apps and websites to validate and refine their mappings.
• Community corrections. Providers like MaxMind and InfoSniper accept correction reports when users identify inaccurate mappings.

Update frequency

Commercial databases are updated on varying schedules:

• MaxMind GeoIP2: Database updates released twice weekly for commercial products.
• DB-IP: Monthly updates for the free tier; more frequent for commercial.
• IP2Location: Monthly updates for commercial products; less frequent for LITE.

The lag between an ISP reallocating IP addresses and the geolocation database reflecting the change is typically 1-4 weeks for major providers. Edge cases — particularly when small ISPs get acquired or merge networks — can persist for months.

# MaxMind correction form:
https://www.maxmind.com/en/geoip-location-correction

# IP2Location correction:
https://www.ip2location.com/register-for-update

# InfoSniper correction:
https://www.infosniper.net/geoip-data-correction.php

If you manage IP ranges, publish a GeoFeed per RFC 8805 to
ensure all providers can access your geographic assignments.

When Accuracy Matters Most — and What Level You Actually Need

Different applications have different accuracy requirements. Mismatching the resolution you need to the resolution the data provides leads to either over-engineering or poor decisions.

Fraud detection: how accuracy affects decision-making

In fraud scoring, a geolocation mismatch between the billing address and the IP-derived location is a risk signal, not a verdict. A credit card billing address in Chicago paired with an IP mapping to Lagos is a strong signal. The same card with an IP mapping to Milwaukee (50 miles away) is noise — that could easily be CGNAT or a cell tower routing to a neighboring city.

The practical approach is to treat IP geolocation as one weighted input in a scoring model. Use country-level data with high confidence. Use city-level data to flag anomalies for review rather than for automated blocking. Always check whether the IP belongs to a known VPN or proxy service — that metadata is often more actionable than the geographic coordinates.

Compliance: when country-level is enough (and when it is not)

For GDPR jurisdictional checks or OFAC sanctions screening, country-level resolution is typically sufficient and reliable enough. If you need to determine whether a user is in the EU or in a sanctioned country, IP geolocation is a reasonable first-pass filter at 99%+ accuracy.

The exception is border regions. A user physically in Windsor, Ontario (Canada) might have their IP routed through Detroit infrastructure and show as being in the United States. For compliance-critical decisions in border areas, supplement IP data with user-provided information.

How to Get Better Results from IP Geolocation Lookups

If you need to maximize the accuracy of your IP geolocation data, these practical strategies help.

1. Cross-reference multiple databases

Academic research consistently shows that using multiple geolocation databases together improves both accuracy and coverage. If MaxMind, DB-IP, and IP2Location all agree that an IP is in Berlin, your confidence is high. If they disagree, the country and region are likely correct but the city assignment is uncertain.

Tools like InfoSniper do this cross-referencing for you, combining multiple data sources to provide a consolidated result. You can also use our API for programmatic access when you need to process lookups at scale.

2. Use the ISP and AS data, not just coordinates

The ISP name and Autonomous System (AS) number returned with a geolocation lookup are often more reliable and more useful than the latitude/longitude. Knowing that an IP belongs to "Comcast Cable Communications" versus "Amazon Technologies" versus "NordVPN" immediately categorizes the connection type and tells you how much to trust the geographic coordinates.

The WHOIS lookup provides additional registration details that can help validate geolocation results, particularly for commercial and data center IPs.

3. Distinguish IP types before interpreting results

Residential broadband IPs are the most accurately geolocated. Mobile IPs are less accurate. Data center and cloud IPs tell you server location, not user location. VPN IPs tell you exit node location. Each type requires different interpretation:

• Residential broadband: The geolocation is probably correct at the city level in developed countries. This is the best case.
• Mobile carrier: Country is correct; city might be off by one metro area. Check if CGNAT is involved.
• Cloud/hosting: The location is the data center, not the end user. Use the ISP field to identify this case.
• VPN/proxy: The location is the VPN server. The reputation checker can identify known VPN providers.
• Satellite (e.g., Starlink): Location is the ground station. Can be hundreds of kilometers from the user.

4. Consider the confidence radius

MaxMind's GeoIP2 database includes an "accuracy radius" field that indicates the confidence level for each specific lookup. A result with an accuracy radius of 10 kilometers is much more reliable than one showing 500 kilometers. If your database provides this metadata, use it — it is the provider's own assessment of how much to trust the specific result.

5. Build for graceful degradation

Design your systems to work at multiple resolution levels. Use country-level data (reliable) for your primary logic. Use city-level data (moderate reliability) for optional enhancements. Never make hard decisions based on postal code or coordinate precision from IP geolocation alone.

// Tier 1: Country-level decisions (high confidence)
if (ip_country === "sanctioned_country") → block

// Tier 2: Region-level enrichment (moderate confidence)
if (ip_region !== billing_region) → flag_for_review

// Tier 3: City-level signal (directional only)
if (ip_city !== billing_city) → add_risk_score_weight

// Never: hard block based on city or postal mismatch alone