Skip to content
Fraud Prevention

IP Fraud Score: What It Means, Score Ranges & How to Check

What an IP fraud score means, what counts as a good one, why an IP gets flagged, and real lookups for a Tor exit, a VPN, a relay and a cloud server.

ByGeoIPHub TeamPublishedUpdatedReading time10 min read
Luminous gauge arc with a single needle above a lattice of nodes, representing an IP fraud score from 0 to 100
Four nodes connected to different points on a gauge, representing four real IPs and their scores
One score, four real IPs, and the reasons behind each.

One score, four real IPs, and the reasons

Four kinds of network land in four different places on the scale, and scoring.detection_methods names every signal that put them there. We looked up these public addresses on 2026-10-16 with the same keyless API call anyone can make:

185.220.101.1

  • What it is: Tor exit (TORSERVERS-NET, Hamburg)
  • Score: 100
  • Action: block
  • detection_methods: tor_exit, datacenter_blocklisted, no_ptr_datacenter, known_spammer

45.83.220.7

  • What it is: Mullvad VPN network address (31173 Services, Malmö)
  • Score: 46
  • Action: review
  • detection_methods: known_vpn_server, datacenter_ip

104.28.46.99

  • What it is: iCloud Private Relay egress (Cloudflare)
  • Score: 9
  • Action: allow
  • detection_methods: relay_ip

45.33.32.156

  • What it is: scanme.nmap.org, Nmap's public test host (Linode)
  • Score: 10
  • Action: allow
  • detection_methods: datacenter_ip

Each signal carries a number of points and a confidence. The score is the sum of points times confidence, rounded and held between 0 and 100. That means you can rebuild every entry above from its methods list:

  • The Tor exit: 100, block. tor_exit (45 points at 0.99), known_spammer (40 at 0.9) and datacenter_blocklisted (30 at 0.75) already add up to 103. The fourth signal only pushes further past the ceiling.
  • The VPN address: 46, review. known_vpn_server (40 at 0.9) plus datacenter_ip (20 at 0.5) is exactly 46. The API names it "Mullvad (31173 Services)" from the network it sits on, although it is not on Mullvad's published server list today. A named VPN is not a high score by itself. Nothing on this server was blocklisted and no open VPN protocol was recorded for it, so it stops at review.
  • The relay: 9, allow. relay_ip adds 10 points at 0.85, which rounds to 9. Private Relay hides an Apple user's address for privacy, and relay addresses are capped at 60, so they can never reach block. We cover how to detect Private Relay and WARP traffic and why to let it through.
  • The cloud server: 10, allow. datacenter_ip alone is 20 points at 0.5. Hosting is a reason to look, not a reason to block, because APIs, uptime monitors and office gateways all run from data centers.

Every reason and its weight is in our field-by-field reason list.

Running these taught us two things the bands above can't. First, the Tor exit and the VPN both hide their users, yet they sit 54 points apart, because the Tor exit also carries spam and blocklist history. The same provider can swing the other way too: Mullvad's listed Frankfurt server, 185.213.155.73, scored 100 on the same day, because its entry on Mullvad's list marks it as a WireGuard server (wireguard_open), and that protocol tag also turns the hosting signal into the stronger datacenter-with-VPN one. Second, one of the Tor exit's four reasons is our gap, not the exit's. no_ptr_datacenter says the address has no reverse DNS, but dig -x 185.220.101.1 returns a Tor exit hostname today. Our lookup database is missing that stored record, and we have logged it for a fix. The verdict does not change, since the other three signals already pass 100. It is still the reason we tell customers to write rules on the signals they trust, not on the raw count.

Both Mullvad numbers have moved since. After our 10 November 2026 scoring change, a VPN with no abuse, spam, botnet, Tor, proxy or honeypot evidence stops at 70, and a hosting address flagged as a VPN gets the datacenter-with-VPN signal. At 06:19 UTC on 11 November, 45.83.220.7 returned 70, step up, with known_vpn_server and datacenter_with_vpn_proxy, and the Frankfurt server returned 70, step up, with the same four reasons as before.

What is a good IP fraud score?

A good score is 25 or under: no meaningful risk evidence, and the API returns allow. Our four bands map straight to actions:

0-25

  • Verdict: Good: no meaningful risk evidence
  • Recommended action: Allow

26-50

  • Verdict: Watch: some signals fired
  • Recommended action: Review: log it, monitor it

51-75

  • Verdict: Risky: material evidence
  • Recommended action: Step up: MFA, CAPTCHA, or 3-D Secure

76-100

  • Verdict: Bad: strong, multi-signal evidence
  • Recommended action: Block

Two caveats stop most mistakes before they happen:

  • Read a score against its own vendor's scale. IPQualityScore's documentation calls 75 and above "suspicious, but not necessarily fraudulent" and recommends flagging or blocking at 90 and above. Our block band starts at 76. A 76 from each of us means different things, so never port a threshold between providers unchanged.
  • The right cutoff depends on the action being protected. A login can tolerate a higher block threshold than a checkout, because a re-prompted password costs seconds while a fraudulent card-not-present order costs a chargeback (the chargeback reduction playbook covers that side). We set out block thresholds for login, signup and checkout one action at a time.

What pushes the number up or down

The score answers one narrow question: how much risk evidence surrounds the network this request came from? It never knows who is behind the keyboard. It knows things like these:

Anonymization

  • Examples: VPN exit, datacenter proxy, Tor exit
  • Effect on the score: Raises: the infrastructure most fraud runs through

Abuse history

  • Examples: Spam sources, credential-stuffing origins, botnet membership
  • Effect on the score: Raises: evidence the address has attacked before

Infrastructure type

  • Examples: Datacenter ASN vs consumer ISP, hosting ranges
  • Effect on the score: Raises for server ranges, since real customers browse from consumer networks

Clean-origin evidence

  • Examples: Residential ISP, mobile carrier, verified search-engine crawler
  • Effect on the score: Lowers: scoring should subtract as well as add

Shared-address safeguards

  • Examples: CGNAT and privacy-relay caps
  • Effect on the score: Limits: one bad user on a shared address cannot condemn thousands

The last two groups separate careful scoring from crude scoring. A model that only adds points ends up flagging every mobile carrier and shared connection in the world. Ours combines 40 weighted reasons and also subtracts: a residential ISP reason takes off 17 points (minus 20 at 0.85 confidence) and a verified crawler 28.5 (minus 30 at 0.95). Relay addresses are capped at 60, and CGNAT addresses without threat evidence of their own at 50. Every reason's weight, and how to read it in a response, is in risk score reasons, field by field, so this page stays on what the number means for you.

Why is my IP's fraud score high?

A high score on your own connection usually comes from the network you are on, not anything you did: a VPN or proxy, an address recycled from an abusive subscriber, a shared CGNAT address that someone else got listed on, or a recent blocklist listing on your ISP's range. An address we recognize as CGNAT is capped at 50 unless it carries VPN, proxy, Tor, botnet or abuse evidence of its own, so a shared address can be reviewed but never blocked on the IP alone. If a site told you that you look like a VPN rather than showing a score, here is why your IP is flagged as a VPN or proxy and what you can do about it.

How to bring a high score down

You cannot appeal the number itself. It moves when the underlying evidence changes, and these steps change it:

  1. Turn off the VPN or proxy and reconnect from your normal ISP line. This is the single biggest and fastest reduction.
  2. Get a fresh address: renew your DHCP lease or reboot your router. On many ISPs this rotates you off a flagged address.
  3. Scan your devices. If something on the network is botnet-infected, the flags will keep coming back until it's cleaned.
  4. Ask your ISP for a different allocation if your range is chronically listed.
  5. For businesses: send transactional traffic from dedicated, properly configured infrastructure with clean reverse DNS, not from shared hosting ranges.

Then allow time. Our VPN probe step runs at most every 12 hours and re-checks findings older than 7 days, our proxy checks run at most every 6 hours and re-check findings older than 2 days, the Tor and relay lists and DNS blocklists are re-read every 6 hours, and the lookup database is rebuilt daily, so when an address drops out of that evidence, its number follows on the next rebuild.

Check a score by API

For a one-off, the free checker linked at the top of this page shows the 0-100 number, the recommended action and the signals behind it. For production, the same verdict comes from one call, and you can try it without a key at a low rate limit. Here is the trimmed real response for the Mullvad network address from the 2026-10-16 list above:

Terminal
curl https://api.geoiphub.com/v1/lookup/45.83.220.7 \ -H "x-api-key: YOUR_API_KEY"
JSON
{ "ip": "45.83.220.7", "asn": { "asn": 39351, "org": "ESAB-AS - 31173 Services AB" }, "detection": { "is_vpn": true, "is_tor": false, "is_hosting": true, "is_relay": false, "vpn_provider": "Mullvad (31173 Services)", "vpn_confidence": 85 }, "scoring": { "fraud_score": 46, "confidence": 0.56, "recommended_action": "review", "detection_methods": ["known_vpn_server", "datacenter_ip"] } }

The provider name comes from our table of networks known to run VPN servers, which is why vpn_confidence was 85 rather than the 95 an address on Mullvad's own server list gets. Since 10 November 2026 that path gives 75, and addresses we have not re-checked since, this one included on 11 November, still show 85. We explain how we name the VPN provider behind an IP and why code should match names by prefix.

That response shape is also a checklist for judging any provider: a number you can act on, a recommended action you can map to policy, a confidence value and, most important, the named evidence behind the verdict. A bare number with no detection_methods is impossible to audit when a real customer asks why they were blocked.

Score at decision points rather than on every pageview. At login, the score is a first filter against credential stuffing and account takeover. At signup, it helps stop fake signups by IP type and catch duplicate accounts from one operator and coupon and referral abuse. At checkout and withdrawal, it is one layer in a payment fraud detection stack. The same network signals also help spot click fraud by IP address on paid campaigns. Pair the IP layer with device and behavioral signals for full coverage, since device fingerprinting and IP intelligence see different things. Our fraud prevention guide shows where the score plugs into each flow, and how VPN and proxy detection works explains where the heaviest-weighted signals come from.

Frequently Asked Questions

Does a high fraud score mean fraud was committed?

No. The score measures the network a request came from (anonymizers, abuse history, infrastructure type), not the intent of the person using it. A high number means the address deserves scrutiny or a challenge, not that its user is a fraudster. That is why mature systems pair the score with step-up verification instead of silent hard blocks.

Does using a VPN raise my score?

Usually. The known-VPN signal alone adds 40 points at 0.9 confidence, which lands in review. An address on Mullvad's network that we looked up on 2026-10-16 scored 46, and 70, step up, on 2026-11-11 after our scoring change. A VPN now stops at 70 unless abuse, spam, botnet, Tor or proxy evidence lifts the cap, so it reaches block only with that kind of evidence on top. Turning the VPN off is the fastest way to present a lower-risk address.

How do I check an IP address's risk score?

Paste the address into a free checker to see its 0-100 score and the signals behind it, or call an IP intelligence API at signup, login or checkout. Whichever provider you use, prefer one that returns the contributing evidence. A bare number you cannot audit is hard to act on and impossible to explain to a blocked customer.

Written by

GeoIPHub Team

The team behind the GeoIPHub API

Written by the GeoIPHub team, the people who build the GeoIPHub IP intelligence API and its scanner. Drafts are prepared with AI assistance and edited by the team; API examples are real lookups against the live API unless labelled illustrative.

Share this article
More articles

Related posts