We have never found a VPN speed claim that was outright false. We have also rarely found one that was useful, and the gap between those two statements is the entire subject of this article.

The four things a speed claim omits

  • The server. Load varies by 61% between endpoints on the same provider in the same city, which is more than any other variable we measure.
  • The hour. We measured a 22% swing between 04:00 and 21:00 on the same server. Marketing numbers are not gathered at 9pm.
  • The baseline. "500 Mbps" means nothing without knowing whether the unprotected line was 550 or 1,000.
  • The sample. A single run is noise. We take a minimum of five per server and report the median.

How the same provider produces two very different numbers

Take our own July results. NordVPN measured 902 Mbps on a nearby server — 96% of our 942 Mbps control line. The same provider on a Frankfurt-to-Sydney route measured 698 Mbps. Both figures are accurate. Publishing only the first would be technically honest and substantially misleading.

Charts on a computer display
Same provider, same day, same test rig: 902 Mbps or 698 Mbps depending entirely on which server you name.

Why "99% of your speed" claims appear

If your line is 100 Mbps, almost every modern VPN will return 95 Mbps or better on a nearby WireGuard server, because the bottleneck is your line rather than the tunnel. Test the same provider on a gigabit connection and the differences appear immediately. Claims of near-zero speed loss are usually true and usually measured on connections too slow to reveal anything.

Any speed figure without a server, an hour and a baseline attached is a marketing number, including ours if we ever forget to publish all three.

How to read any speed claim

  • Ask what the unprotected baseline was. No baseline, no meaning.
  • Ask which distance band. Nearby figures cluster and barely separate providers; long-haul results spread out immediately.
  • Ask how many runs and whether the figure is a median or a peak. Peaks are cherry-picked by definition.
  • Ask what time of day. Peak-hour numbers are the ones that reflect your actual evening experience.

What we publish

Every figure on this site carries the server band, the hour and the baseline. Our raw per-server CSV exports go to anyone who emails and asks, free and without conditions. Several readers have re-run our numbers and sent corrections, three of which changed a published score — which is the entire reason we hand them out.