Accuracy radius now calibrated to a 68% target (November 9, 2026)

From November 9, the radius field in IPinfo geolocation data means one specific thing: there is a 68% chance the IP’s user is within that many kilometers of the reported location. We’ve checked that against where users actually are, and the radius now keeps that promise in every bucket. Coordinates, city, region and country don’t change, and neither do the field’s name, type or buckets.

What the radius tells you

The radius says how spread out an IP’s users are around the point we report. It doesn’t grade how good our data is. The two are easy to confuse, so consider two IPs that both have the correct location:

  • A home fiber line in Lyon. Everyone using it is in one apartment, so the radius is small.
  • A mobile carrier IP in Texas. Thousands of phones share it and it gets reassigned across the state, so the radius is large.

Both locations are right. Only the first has its users close to the point. A large radius is the data telling you something true about the IP: the location represents a region, not a place.

The radius is published as the smallest of these buckets that reaches 68%: 5, 10, 20, 50, 100, 200, 500, 1000, 2000 and 5000 km. A 5000 km radius means the IP’s users are spread across a continent, so don’t rely on anything finer than the country.

Why 68%?

68% is the standard way to state location accuracy. Phones use it for GPS: Android defines horizontal accuracy as the radius of 68% confidence, the blue circle around your position in a map app. An IP’s radius now reads the same way.

A 100% radius sounds safer, but it would tell you less. Any IP has some users who show up far from the rest: travelers, people on VPNs or corporate tunnels, phones roaming on a foreign SIM. A radius that had to include all of them would be set by those few. Most IPs would get radii of thousands of kilometers, and you could no longer tell a precise IP from a vague one. A 95% target has the same problem, just less of it.

At 68%, the radius describes where most users of an IP actually are, and that’s the information you need to tell IPs apart. It also means about one user in three is outside the radius by design. That isn’t an error. If your use case needs more certainty than that, build in margin instead of treating the radius as a hard boundary.

Why we can make that promise

To promise 68%, you need to know where an IP’s users actually are, and how close the IP can physically be to the point you report. Without both, a radius only says how confident the provider feels. IPinfo has both:

  • Where users are. Location data from real devices shows how spread out each IP’s users really are. We measure the radius instead of assuming it, and that same data is how we check that the promise holds.
  • Where the IP can be. ProbeNet, our network of 1,400+ measurement points, times round trips to IPs around the world. A signal can only travel so far in a few milliseconds, so a fast round trip proves an IP is close. That lets us give tight radii where we’ve measured, including servers and hosting ranges that no phone ever uses.

Each source covers the other’s blind spot. Measurements show how near an IP is. Devices show when its users still end up far away, for example when mobile traffic is routed through a gateway hundreds of kilometers from the phone.

We tested the new radius on networks it never learned from, against where users actually were the next day. Up to 200 km, where most IPs sit, every bucket holds within a few points of 68%.

What changes for you

Only the radius values change. Coordinates, city, region and country stay the same, and so do the field’s name, type and buckets. You’ll see some values move:

  • Mobile carrier ranges get larger radii. Their users really are spread out, and the old radius understated it.
  • Fixed-line ISP ranges barely move, and many get tighter.
  • Hosting and data center ranges stay small where our measurements pin them down.

If you filter on the radius, review your cutoffs, especially around 200 km for mobile traffic. More mobile IPs will cross it, and they should. If a model uses the radius as a feature, retrain it on the new values.

For a before/after on your own IPs, contact support or your account manager.