Skip to content

How GeoIPHub classifies an IP address.

Every flag traces back to a method, a source and a timestamp. Here are the seven methods, how often each source refreshes, and how the evidence becomes a score.

Verified over aggregated.

Most IP data merges vendor lists that disagree, decay and never say why. We set a flag only when we can verify it, and the evidence ships in the response.

  • A protocol that answered, a provider's own list, a registry record or an official feed
  • scoring.detection_methods names every signal behind a verdict
  • A field we can't substantiate returns "none", never a guess

Seven method families.

No single method decides. Each adds evidence, and direct verification carries the most weight.

ASN classification

853 curated hosting ASNs, known VPN networks, and satellite, anycast and CDN tables, matched against the global routing table, which refreshes every 12 hours.

Official provider lists

VPN providers' own server lists plus a public third-party list, and official IP ranges from 10 cloud providers.

Active protocol probing

Real handshakes on 25+ ports across 8 VPN and proxy protocols. A server that answers OpenVPN is running OpenVPN.

Threat and blocklist feeds

Spamhaus DROP/EDROP, FireHOL Level 1, Feodo Tracker and bogons, plus 3 DNS blocklist zones and known scanner ranges.

WHOIS and RDAP

Records from all 5 RIRs, checked for 46 VPN and 27 hosting keywords. Supporting evidence only, never the sole reason to flag.

Verified crawlers

A User-Agent is a claim. Crawlers and AI bots count as verified only inside their operators' 19 official IP-range feeds.

Live classification

An IP we haven't seen is classified live in under 2.5 seconds, then stored for every lookup after that.

See it on a real IP

Run a lookup and read the detection methods that fired for yourself.

Refresh cadence.

A botnet list goes stale in days, a WHOIS record in months, so each source refreshes on its own schedule.

Threat feeds, Tor exit and relay lists, DNS blocklists
Every 6 hours
VPN and proxy probes (hits re-checked after 7 days), BGP routing table
Every 12 hours
Crawler and AI-bot ranges, cloud provider ranges, lookup database rebuild
Daily
WHOIS registry files from the 5 RIRs
1 to 2 days

How evidence becomes a score.

The 0 to 100 fraud_score weighs 40 weighted reasons, and direct verification counts most.

  • It subtracts too: a residential ISP takes 17 points off, a verified crawler 28.5
  • Shared CGNAT ranges are capped, so one bad neighbour can't condemn thousands of users
  • allow 0–25 · review 26–50 · step_up 51–75 · block 76–100

Architecture in brief.

Deliberately boring, so lookups stay fast and every answer keeps its evidence.

  • A Rust scanner runs 15 scheduled steps: feeds, lists, probes, RDAP, DNS and classification
  • Postgres holds every classification and its evidence trail
  • A memory-mapped database, swapped atomically once a day, serves sub-millisecond reads

Start checking IPs in minutes

Create a free API key and get 1,500 lookups a day with every data field included. Upgrade only when your traffic grows.

  • 1,500 free lookups a day
  • No credit card
  • Every field on every plan