How we test and rate VPNs
Every score on this site comes from the same tests, on the same hardware, weighted the same way. This page is the whole method — including what we can't measure.
Last updated September 4, 2026
Every figure this method produces is published — the reviews carry the per-VPN numbers, and the best-for pages re-rank the same data by job.
The five criteria and their weights
A VPN's overall rating is a weighted average of five sub-scores. The weights are fixed for the calendar year, published here, and identical for every VPN. We do not adjust them per provider, and a commission never enters the calculation.
| Criterion | Weight | What we measure |
|---|---|---|
| Privacy & security | 30% | Logging policy, audits, jurisdiction, leak tests |
| Speed | 25% | Download, upload and latency in four regions |
| Streaming | 15% | Netflix, Prime, Disney+, BBC, DAZN |
| Apps & ease | 15% | Every platform, kill switch, split tunnelling |
| Value | 15% | Real 2-year price, devices, refund terms |
The test rig
All speed tests run on a 940 Mbps fibre connection using Windows 11 desktop, MacBook Air M3, iPhone 16, Pixel 9, Fire TV Stick 4K. We test servers in US East, UK, Singapore, Australia — 3 runs per server, median reported. The baseline (no VPN) is measured immediately before each VPN run so a slow evening on the line doesn't penalise one provider.
We buy every subscription with our own money at the public price. Providers do not get early sight of results and cannot request changes.
Privacy & security (30%)
We read the logging policy and the most recent independent audit, check the company's jurisdiction and ownership, and run DNS, IPv6 and WebRTC leak tests on Windows, macOS, iOS and Android with the kill switch on and off. RAM-only servers, audited no-logs claims and a history of behaving well under legal pressure raise the score; a logging incident, an unaudited claim or a missing kill switch lowers it.
Speed (25%)
Download, upload and latency to each region, three runs per server, median reported. The sub-score is based on the share of baseline bandwidth retained on average across the four regions, with a penalty for large long-distance drop-offs and for added latency above 15 ms on a nearby server.
Streaming (15%)
We try to play a title on 14 services — Netflix US, Netflix UK, Netflix JP, Netflix CA, Netflix AU, Prime Video, Disney+, Hulu, Max, Paramount+, Peacock, BBC iPlayer, ITVX, DAZN — from at least two servers each. A service counts as unblocked only if it plays on the first or second server without a proxy error. The matrix on each review is re-run monthly and dated.
Apps & ease (15%)
Platform coverage, whether the kill switch is on by default, split tunnelling, obfuscation, the quality of the TV apps, and how many steps it takes to connect for the first time. We also count simultaneous connections.
Value (15%)
The real cost over two years including what the plan renews at, device limits, refund length, and whether extras are bundled or upsold. We claim every money-back guarantee ourselves and record how many days the refund took.
What we can't measure
We cannot see inside a provider's servers. A no-logs audit is the best available evidence, not proof, and we say so on every review. We also cannot test every server: a streaming result means "worked on the servers we tried this month", not a guarantee.
Re-testing and changes
Every VPN is re-run monthly. The next scheduled re-test is October 2, 2026. Score changes are listed on the home page changelog with the reason, and each review shows the date it was last tested.
Money
We earn a commission when you buy through links on this site. That pays for the subscriptions and the hardware. It has no effect on scores or rank order, and a VPN that pays us nothing (Mullvad, for example) is rated exactly like one that does. Full details on the affiliate disclosure page.