When researchers study the global internet, they frequently rely on Virtual Private Networks (VPNs) to collect measurements from different economies and regions. A fundamental assumption has long underpinned much of this empirical research: that a VPN connection originating from a specific economy will provide a roughly standardized, homogeneous view of the web. Whether a scientist uses one commercial provider or another to simulate a connection from Chile, Italy, or the United States, the underlying assumption has been that the resulting data will tell the same story about digital infrastructure and data flows.
A comprehensive new study challenges that core assumption, revealing that the choice of VPN provider can dramatically alter what researchers observe online. According to the research, switching between different commercial VPN services can fundamentally change the specific servers a browser connects to, as well as the economies where those servers are physically located—even when the connection originates from the exact same city and economy. These provider-dependent variances can lead academic studies and policy assessments to significantly overestimate or underestimate how much web traffic remains within national borders, how much crosses international jurisdictions, and which economies are truly hosting the critical infrastructure serving everyday users.
The implications of these findings extend far beyond academic curiosity. As governments around the world increasingly debate and implement strict data localization laws, cross-border privacy regulations, and digital sovereignty mandates, the tools used to measure the internet carry immense weight. A study relying on one particular VPN provider might conclude that a specific nation’s web traffic is overwhelmingly served by domestic infrastructure, while a parallel study using a different provider might indicate that the very same nation is heavily reliant on foreign servers. Consequently, the choice of measurement tool shapes not only the raw data collected, but also the high-stakes policy conclusions drawn regarding national security, data exposure, and the geographic distribution of the modern internet.
How researchers tested whether VPN choice changes Internet measurements
To investigate the true impact of VPN selection on internet telemetry, the study’s authors developed a specialized system known as VPN-Scope. This measurement framework was designed to visit identical sets of websites through multiple distinct VPN providers at nearly simultaneous times, utilizing connections routed through the exact same economy and, whenever possible, the exact same municipality.
The empirical investigation spanned four major commercial VPN providers—ExpressVPN, NordVPN, ProtonVPN, and Surfshark—and encompassed 14 distinct economies spread across Europe and the Americas. To ensure a representative sample of web traffic, the researchers selected 1,000 websites per economy, blending globally dominant web platforms with locally popular portals and official government websites. Each target website was scheduled to receive 25 visits through each of the four participating providers over a rigorous three-week observation window in March and April 2025.
During every automated visit, the system meticulously recorded all servers contacted during the page-loading process, including third-party infrastructure responsible for supplying embedded resources such as stylesheets, scripts, and fonts. Researchers then analyzed the geographical locations and corporate hosting organizations associated with each encountered server. Supplemental experiments were also conducted to scrutinize Domain Name System (DNS) resolution behaviors and isolate the physical locations serving content through the various VPN tunnels. Together, these multi-layered tests allowed the research team to connect the observed discrepancies to the underlying technical mechanisms driving them.
How much VPN providers changed the measured Internet infrastructure
The empirical results underscored profound discrepancies in how different VPN services interact with global internet routing. Data gathered from measurements originating in Chile vividly illustrated how VPN selection alters the perceived ratio of domestic versus international traffic. When requests were routed through ExpressVPN from Chile, just 9.5% of the analyzed requests resolved to server addresses geolocated within Chilean borders. By contrast, the domestic traffic share skyrocketed to 62.7% when using Surfshark and 64.2% when using NordVPN.
The variance extended deeply into neighboring regions as well. For the same set of Chilean web requests, Brazil accounted for 57.1% of destination endpoints when measured via ExpressVPN, yet dropped precipitously to just 2.8% through Surfshark and 2.4% through NordVPN. Because these measurements captured identical website lists from the same source economy, the divergence proved that the choice of VPN provider is far from a trivial implementation detail. Instead, switching providers can actively alter the fundamental geographic conclusions a researcher derives from otherwise comparable data sets.
For policymakers and researchers evaluating data-localization policies, such discrepancies introduce significant analytical challenges. An investigator utilizing NordVPN would observe a strong majority of requests tied to domestic servers, whereas an analyst using ExpressVPN would record the exact opposite conclusion. While these network-level measurements do not definitively track where user data is permanently stored at rest, they drastically change the apparent reliance on domestic versus foreign infrastructure.

Similar patterns emerged across other tested regions. Measurements originating from the United States and the United Kingdom showed that while major destination hubs like the US or Western European nations remained prominent across most providers, the precise distribution percentages shifted notably depending on the commercial routing paths utilized.
Furthermore, DNS resolution experiments revealed that when comparing two distinct providers within the same economy, the set of returned IP addresses differed in roughly 12% of comparisons for primary website domains. When examining dependency domains—such as the auxiliary services supplying scripts, fonts, and analytics—the rate of disagreement surged to approximately 31%. Relying solely on the initial domain name typed into a browser risks missing a substantial portion of the structural variation occurring beneath the surface.
Despite these differing IP addresses, the underlying corporate entities hosting the content displayed remarkable stability. At the organization level, DNS answers from different VPNs shared at least one common hosting organization in 98% of initial-domain comparisons and 96% of dependency-domain comparisons. This high degree of corporate overlap indicates that while VPN selection heavily distorts the perceived geographic location of web infrastructure, it has a lesser impact on identifying which major technology and hosting companies are ultimately responsible for delivering the content.
What researchers can control to make VPN measurements more reliable
While certain sources of variation are inherent to commercial routing, researchers can implement rigorous experimental controls to minimize measurement error. Best practices include independently verifying VPN server exit locations, standardizing starting cities whenever feasible, and maintaining consistent browser configurations and precise measurement timing. Furthermore, scientists are urged to transparently document the specific provider, connection protocols, and network configurations utilized, moving away from vague descriptions like a measurement originating simply "from Italy." The specific DNS resolver answering domain queries must also be carefully monitored, as selecting a uniform DNS service in client settings does not guarantee that every VPN provider will actually route queries through it.
Other discrepancies, however, stem from autonomous decisions made by the wider internet ecosystem. Content Delivery Networks (CDNs) continuously distribute replicated copies of web assets across numerous global data centers, dynamically selecting different physical servers based on the originating network path of an incoming request. Network routing policies can similarly steer two separate connections toward entirely different physical facilities, even when both connections share the exact same destination IP address.
Targeted testing demonstrated that VPN connections initiating from Milan frequently terminated at content-serving facilities located either locally in Milan or across borders in Frankfurt. Sharing a starting city and a destination address was frequently insufficient to guarantee a shared physical destination. Because these routing decisions rely on complex commercial peering and traffic-engineering arrangements managed by network operators, researchers cannot fully standardize them through client-side browser or VPN settings alone.
Ultimately, these findings suggest that the VPN provider should be treated as an integral component of experimental design rather than a passive utility for obtaining an overseas connection. When geographic accuracy is paramount, researchers should report their exact provider settings, independently verify exit node locations, and consider replicating measurements across multiple providers to account for routing biases.
As Sachin Kumar Singh, a postdoctoral researcher at the University of California specializing in security, privacy, and internet measurement, notes, conclusions regarding endpoint economies, traffic localization, and physical server distribution remain highly sensitive to VPN selection. For future studies examining cross-border data flows and internet infrastructure geography, data gathered through a single VPN provider should be interpreted as one distinct perspective of the network rather than an absolute, provider-independent representation of the global economy.