On IP Geolocation Accuracy: A Comparative Study

This study compares IP geolocation accuracy across ten major API providers using ground truth GPS coordinates collected from real website visitors via the browser's Geolocation API.

Abstract

IP geolocation is fundamental to content localization, fraud detection, and regulatory compliance, yet the accuracy of commercial services varies significantly. This study evaluates ten providers against ground truth GPS coordinates collected with user consent. After filtering non-residential IPs (datacenters, VPNs, proxies), providers show city-level correct rates (≤50 km deviation) ranging from ~50% to ~75%, while stricter neighborhood-level accuracy (≤10 km) drops to ~15% to ~35%. The 75th percentile deviations range from ~128–288 km.

Methodology

Data Collection

When visitors navigate to ipapi.is, a permission dialog requests geolocation access. Upon consent, we record GPS coordinates, accuracy radius, timestamp, IP address, timezone, and device metadata.

Ground Truth: Browser Geolocation API

The browser's Geolocation API provides ground truth coordinates. Mobile devices with GPS typically report 5–15m accuracy, while desktop browsers using WiFi triangulation report 35–100m. Values above 1,000m indicate unreliable IP-based fallback and are excluded from analysis.

IP Filtering

Non-residential IP addresses are excluded: datacenters (server traffic unrelated to user location), VPNs/proxies (traffic routed through intermediate servers), Tor (exit nodes), abusers (blocklisted IPs), satellite (serves vast areas), and crawlers. Only IPs where all ipapi.is classification flags are false are included.

Additionally, we use external verification sources such as spur.us to identify and exclude residential proxies that may evade standard detection methods. IPs where all geolocation providers unanimously report the same incorrect country (with deviations exceeding 1,000 km) are also flagged as likely anonymized connections and removed from analysis.

We also apply cross-provider country consensus filtering: if 70% or more of the evaluated geolocation providers agree that an IP address is in the same country, and that consensus country differs from the user's GPS ground truth country, the IP is flagged as a suspected residential proxy and excluded from analysis. For example, if a user's GPS shows Lagos, Nigeria but 8 out of 10 providers agree the IP is in the United States, this strongly suggests the user is routing traffic through a US-based residential proxy.

Evaluated Providers

Ten IP geolocation providers were evaluated:

• ipapi.is ⭐ this API
• ipinfo.io
• ipdata.co
• ip-api.com
• iplocate.io
• ipwho.is
• ipgeolocation.io
• freeipapi.com
• ipbase.com
• db-ip.com

Accuracy Metrics

Distance Deviation: Haversine distance between GPS ground truth and provider coordinates. Correct/Wrong Classification: We use two thresholds—a guess is "correct at 50 km" if ≤50 km from true location (city-level accuracy), and "correct at 10 km" if ≤10 km (neighborhood-level accuracy). Country Accuracy: Whether the provider returns the correct country code compared to reverse-geocoded ground truth.

Results

Dataset Overview

Accuracy Statistics by Provider

Deviation statistics for each provider based on clean residential IP addresses (distances in km):

Analysis

Sample Comparison

Individual IP lookups showing accuracy variation across providers:

Country Mismatch Examples

Cases where providers incorrectly identified the country, demonstrating IP geolocation limitations:

Discussion

Limitations

• Sample Size: Relatively small; results may not generalize to all regions or network types.
• Geographic Bias: Sample skews toward regions where ipapi.is is better known.
• Ground Truth Accuracy: Browser geolocation varies by device (35–100m for WiFi, 5–15m for GPS).
• Temporal Factors: Databases update at different frequencies; results are a snapshot in time.

Why Filtering Matters

Excluding VPN/proxy traffic is critical. A user in Berlin using an Amsterdam VPN would show ~650 km deviation even though the provider correctly identifies the VPN endpoint. This is correct behavior, not an error.

Outliers

Maximum deviations of ~15,000+ km occur with mobile carrier IPs (geolocated to headquarters), CGNAT addresses (shared across wide areas), IPv6 with limited data, and recently reallocated IP ranges with stale records.

Conclusion

In summary, IP geolocation is far from perfect—but it is also far from random guessing. The data shows that IP geolocation achieves approximately 92% accuracy at the country level and around 70% accuracy at the city level (within 50 km). This makes it a reliable first indicator of where a user is located. For applications requiring higher precision, browser-based geolocation (GPS/WiFi triangulation) should be used as a complement or replacement.