People often think VPN detection is some clever trick that looks inside the encrypted tunnel. It is much simpler than that. The site looks at the address your traffic comes from and asks a database what kind of network owns it. If the answer is a data center, that is usually enough.
We wanted real numbers instead of claims, so we took the server lists that three VPN companies publish themselves and checked every address the way an ordinary website could.
VPN detection at a glance
| What we checked | Result |
|---|---|
| VPN server addresses tested | 833 (Mullvad 530, IVPN 88, Private Internet Access 215) |
| Countries those servers are listed in | 113 |
| Flagged as proxy, VPN or data center by a free database | 824 of 833 (98.9%) |
| Real home connections flagged by the same database | 7 of 400 (1.8%) |
| Tor exit nodes flagged by the same database | 98.7% of 300 |
| VPN addresses on at least one of 15 blocklists | 49.6% |
| Date of the test | 21 November 2026 |
How do websites know I'm using a VPN?
There are five common ways, and only the first one needs nothing from your device.
| Method | What the site checks | Does a VPN hide it? |
|---|---|---|
| Address database | Who owns the network your address belongs to: home provider, mobile carrier or data center | No. This is the method our test measures |
| Known VPN lists | Lists of server addresses gathered by detection companies, sometimes from VPN companies' own published lists | No |
| Many users on one address | Hundreds of accounts arriving from the same address in a day | No. Everyone on that server shares it |
| Your browser's other signals | Time zone, language and keyboard settings that do not match the country of the address | Only if you change those settings too |
| Leaks | WebRTC or DNS requests that reveal your real address or your home provider's DNS | A well set up VPN blocks them. Our VPN leak test checks |
Most sites stop at the first row because it is cheap and needs no code in your browser at all.
What did our test find?
We used ip-api.com, a free IP database that many small sites use. It returns two flags that matter here. Its own documentation describes "proxy" as a "Proxy, VPN or Tor exit address" and "hosting" as "Hosting, colocated or data center". A site that blocks on either flag would treat the address as a VPN or a server.
| VPN service | Addresses tested | Flagged "proxy" | Flagged "hosting" | Flagged either | Not flagged |
|---|---|---|---|---|---|
| Mullvad | 530 | 98.9% | 88.9% | 99.4% | 3 |
| IVPN | 88 | 95.5% | 98.9% | 100.0% | 0 |
| Private Internet Access | 215 | 95.8% | 34.0% | 97.2% | 6 |
| All three | 833 | 97.7% | 75.8% | 98.9% | 9 |
To see whether the database simply flags everything, we ran the same check on two other groups. The first was 400 home connections that volunteers run as RIPE Atlas probes in 60 countries: ordinary broadband at people's houses. The second was 300 Tor exit nodes taken from the Tor Project's public list.
| Group | Addresses | Flagged "proxy" | Flagged "hosting" | Flagged either |
|---|---|---|---|---|
| Home connections (RIPE Atlas) | 400 | 0.3% | 1.5% | 1.8% |
| VPN servers (three providers) | 833 | 97.7% | 75.8% | 98.9% |
| Tor exit nodes | 300 | 93.0% | 86.7% | 98.7% |
So the label is not noise. A home connection is almost never flagged, and a VPN server almost always is. That gap is what makes this the method nearly every site uses.
Where do VPN servers actually run?
The database also tells you whose network each address belongs to. Most VPN companies rent servers from a handful of hosting firms, and those firms' names show up again and again.
| Network shown by the lookup | VPN addresses | Share of 833 |
|---|---|---|
| Datacamp Limited | 202 | 24.2% |
| tzulo, inc. | 136 | 16.3% |
| M247 Europe SRL | 90 | 10.8% |
| GSL Networks Pty LTD | 83 | 10.0% |
| 31173 Services AB | 62 | 7.4% |
| HostRoyale Technologies Pvt Ltd | 37 | 4.4% |
| M247 Europe Infra | 20 | 2.4% |
| Mullvad | 15 | 1.8% |
Four networks carry 61.3% of all the addresses we tested. A detection company that labels those four networks catches most VPN traffic without knowing anything about the VPN brands themselves.
Owning the hardware does not help either. Mullvad's own list marks 110 of its relays as owned rather than rented. All 110 were flagged, against 99.3% of the 420 rented ones.
Do VPN addresses end up on blocklists?
Blocklists are a separate system, built for email, but many sites read them too. We checked every address against the same 15 lists our blacklist check uses, after testing that each list answered correctly.
| Blocklist | VPN addresses listed | Share of 833 |
|---|---|---|
| SPFBL | 303 | 36.4% |
| Spamhaus ZEN (any code) | 146 | 17.5% |
| UCEPROTECT Level 2 | 68 | 8.2% |
| s5h | 20 | 2.4% |
| Barracuda | 17 | 2.0% |
| JustSpam | 3 | 0.4% |
| Mailspike | 2 | 0.2% |
| blocklist.de | 1 | 0.1% |
In total 49.6% of the addresses were on at least one list and 14.4% were on two or more. As in our blocklist study of 1,000 IPs, most Spamhaus hits were the Policy Block List, which only says an address should not send email directly. Without those policy entries and the network wide UCEPROTECT Level 2 listings, 39.3% of the VPN addresses still had a listing of their own, and only 1.7% had a real Spamhaus listing.
For you as a VPN user this matters less than it sounds. The listing belongs to the shared server, and it mostly affects sending email from that address.
Do virtual locations fool geolocation?
Some VPNs offer countries where they have no server. Private Internet Access marks these regions as virtual in its own server list: 94 of its 215 regions. The traffic leaves from a server in another country, but the address is registered to the advertised one.
The free database showed the advertised country for 90 of those 94 virtual regions. Across all 833 addresses, only 1.2% showed a different country from the one the VPN lists. So a virtual location usually looks right to a website. What gives it away is the hosting label, which still caught 98.9% of the virtual regions.
If you want to see what a site sees for any address, paste it into our IP lookup. For why locations go wrong in general, read why your IP location is wrong.
Why weren't 9 servers flagged?
Nine addresses passed the free check: four on Datacamp Limited, three on M247 Europe SRL, one on Micfo and one on WorldStream. Those are the same hosting networks as the flagged ones, so the gap is in this one database's records for those particular ranges, not in anything the VPNs did differently.
What to do when a site blocks your VPN
- Switch to another server in the same country. Blocks are often made range by range, so a different server can work.
- Try a different country only if the site allows it. Streaming and banking sites often refuse foreign addresses on purpose.
- Turn the VPN off for that one site. Many VPN apps have a split tunnel setting for this.
- Check for leaks first. If a site still sees your real country with the VPN on, run our VPN leak test. Our guide to VPN leaks explains the fixes.
- Do not buy a "dedicated IP" expecting it to be invisible. It is still an address in a hosting network, so it is still labelled as hosting. It only stops you sharing the address with strangers.
How we ran the test
- VPN addresses. We downloaded the server lists each company publishes: Mullvad's relay list (every active WireGuard relay), IVPN's server list (every active gateway, resolved to its address) and Private Internet Access's server list (the first WireGuard server in each region, so large regions did not count twice). Proton VPN's list now needs an account, so it is not included.
- Exit addresses. We tested the server addresses each company publishes. A provider can send traffic out through a different address than the one it lists, so these figures describe the published servers.
- Home connections. 400 RIPE Atlas probes tagged as home connections, sampled at random with at most 20 per country, from the RIPE Atlas probe list.
- Tor exits. 300 addresses picked at random from the Tor Project's exit list.
- Detector. ip-api.com's free batch endpoint. Its terms allow non commercial use only, which this research is.
- Blocklists. 15 DNS blocklists, each tested with the standard 127.0.0.2 and 127.0.0.1 check first. 2 of 12,495 lookups timed out and were left out.
All checks ran on 21 November 2026.
Before you pay for a VPN
A VPN changes the address sites see, and that is exactly how they spot it. In our test one free database flagged 98.9% of real VPN server addresses and only 1.8% of home connections, so any site that wants to know can know. What a VPN still does well is keep your traffic away from the local network and your provider, and hide your home address from the sites you visit. Choose it for that, not to look like a person at home.
