Internet measurement research relies heavily on Virtual Private Networks to observe traffic patterns across different regions, yet a fundamental assumption underlying this work may be flawed. When researchers measure web traffic through VPN connections originating from the same country, they typically expect to see similar patterns regardless of which VPN provider they choose. New research challenges this assumption, demonstrating that the selection of VPN provider can substantially alter which servers appear to serve users and where those servers are geographically located.
The implications are significant for studies examining data localization, cross-border traffic flows, and the geographic distribution of Internet infrastructure. A measurement campaign using one VPN provider might conclude that web traffic remains largely within a country's borders, while the same websites measured through a different provider could suggest heavy reliance on foreign infrastructure. This variability means that conclusions drawn from VPN-based measurements may reflect the characteristics of the measurement tool as much as the actual state of the Internet.
Testing VPN provider impact on measurements
To examine whether VPN selection influences measurement outcomes, researchers developed VPN-Scope, a system designed to visit identical website sets through multiple VPN providers at nearly the same time from the same geographic location. The study encompassed four providers—ExpressVPN, NordVPN, ProtonVPN, and Surfshark—across 14 economies in Europe and the Americas. For each economy tested, the researchers selected 1,000 websites combining globally popular sites with locally relevant and government websites. Over a three-week period in March and April 2025, each website received 25 visits through each provider.
During each measurement, researchers recorded all servers contacted as pages loaded, including those delivering embedded resources like scripts and fonts. They documented server locations and hosting organizations, then conducted supplementary experiments to examine Domain Name System behavior and identify the physical locations actually serving content through different VPNs. These additional tests helped establish which mechanisms were responsible for the observed differences.
Dramatic differences in geographic conclusions
Measurements from Chile illustrate how dramatically VPN provider selection can alter the geographic picture of Internet infrastructure. When measuring the same website set from Chile, ExpressVPN showed 9.5% of requests associated with server addresses geolocated within Chile, while Surfshark indicated 62.7% and NordVPN showed 64.2%. The distribution of traffic attributed to other countries shifted equally dramatically: Brazil accounted for 57.1% of requests through ExpressVPN but only 2.4% through NordVPN.
These measurements came from identical source locations and website sets, yet produced substantially different geographic distributions. A researcher employing NordVPN would observe that most analyzed requests connected to domestic servers, while one using ExpressVPN would reach the opposite conclusion. For policy research examining data localization requirements, such a difference could fundamentally alter perceptions of whether an economy depends primarily on domestic or foreign infrastructure, despite the measurements not directly revealing where data is ultimately stored.
Variation also emerged in DNS experiments. When comparing two providers within the same economy, the returned address sets differed in approximately 12% of comparisons for primary website domains. For dependency domains—the additional services supplying scripts, fonts, analytics, and other page components—disagreement rose to approximately 31%. Focusing only on the website address a user types into a browser therefore misses a substantial portion of the variation introduced by VPN choice.
However, different addresses did not consistently indicate different hosting companies. At the organizational level, compared DNS answers shared at least one hosting organization in 98% of initial-domain comparisons and 96% of dependency-domain comparisons. VPN choice therefore affects perceptions of server locations more strongly than perceptions of which companies operate those servers.
Improving reliability through experimental design
Certain sources of measurement variation can be reduced through careful experimental methodology. Researchers can independently confirm the actual locations of VPN servers, conduct measurements from the same city where feasible, and maintain consistent browser settings and timing across tests. Documentation should specify the VPN provider, connection settings, and network used rather than simply stating a measurement came from a particular country. The DNS service responsible for resolving website names also requires verification, since different VPNs may not consistently use the same DNS resolver even when configured identically.
Other sources of variation originate from decisions made by network operators themselves, beyond researcher control. Content Delivery Networks distribute content across multiple locations and select different servers based on the network originating each request. Routing decisions can direct two connections to different physical locations even when both target the same destination address. In targeted experiments, VPN connections originating in Milan reached content-serving locations in either Milan or Frankfurt, demonstrating that a shared starting city and even a shared destination address cannot guarantee identical physical destinations.
Treating VPN provider as an experimental variable
The findings suggest that VPN provider selection should be treated as a deliberate component of experimental design rather than merely as a tool for obtaining a connection from a particular region. Researchers should document which provider and connection settings they employed, verify the actual location of the VPN exit point, and when geographic conclusions are important to their work, consider repeating measurements across multiple providers.
Even with these precautions, complete elimination of variation remains impossible. DNS infrastructure, CDN routing decisions, and network path selection can still cause two VPN connections from the same economy or even the same city to reach different servers or physical locations. The significance of this residual variation depends on the specific research question being addressed.
Conclusions about which economy an endpoint resides in, whether traffic appears to remain domestic, and which physical content replica serves users prove more sensitive to VPN provider choice. Conclusions about which hosting organizations are involved generally remain more stable across different providers. For studies examining cross-border data flows, data localization policies, or the geographic distribution of Internet infrastructure, results obtained through a single VPN provider should be understood as representing one perspective on the network rather than a provider-independent characterization of the economy.