Domain reputation score for 20 million domains
A category says what a domain is. It does not say whether to trust it.
We score every domain on 21 signals, not just its category: popularity, recent registration, hosting neighborhood, name shape, DNS records, malware and phishing reports and more.
Two of them need data at a scale that is hard to match: our database of 1.6 billion SNIs and host names shows the infrastructure each domain runs, and our multi-year phishing history shows which domains were used for phishing before. Each of the 20M domains gets a trust score from 0 to 100, a grade from A to F and short reason codes.
Domain reputation scores are included with the 20M Web Filtering Database; ask for a quote
Every row gets a score, a grade, reason codes and separate content flags. They sit next to the web filtering category, keyed by the same domain.
- Infrastructure footprint: the real host roles each domain runs
- Phishing history: past reports keep counting and fade with time
- 19 more signals: popularity, recent registration, hosting neighborhood, name shape, malware reports and others
Infrastructure footprint: real organisations run real infrastructure
We look at every host name a domain runs and the job of each one, from our database of 1.6 billion SNIs and host names. Real organisations run login, mail, API and CDN hosts, while throwaway and phishing domains run one host or random ones.
Established organisation
Many distinct roles, each with its own hosts. This takes a team and a budget to build.
Throwaway or phishing domain
One host, or many random names with no clear job. Nothing behind the page.
A login system, email, an API for apps and a CDN for traffic each run on their own hosts.
Building and running it all takes months. Throwaway, phishing and parked domains almost never have it.
Each distinct role adds trust. The first roles add the most, later ones a little less.
One host or random names earn no bonus. Phishing history and a parked state still win.
How trust grows with distinct infrastructure roles
Roles we count include login, API, CDN, download, push, telemetry, remote access, network and ad hosts. Mail alone does not count, because cheap hosting adds it to almost every domain.
A broad set of roles can carry a quiet but real organisation to grade B or A.
Phishing history: past reports keep counting
A domain that was ever reported for phishing loses trust, even after it was cleaned. Recent and repeated reports weigh more, older ones fade.
We keep a multi-year history of our phishing feed. Shared platforms and hacked established sites are not counted as phishing domains.
Every reported host name, including pages on shared platforms and on hacked sites.
How the penalty fades over time
Phishing on the domain or its subdomains marks the whole domain. It drops to D or F.
Only the reported page host is marked. The platform itself keeps its score and gets a USER_CONTENT flag.
The site gets PHISHING_ON_SUBDOMAIN and a modest penalty. The reported host itself keeps the full penalty.
Why a category alone is not enough for blocking decisions
A category answers one question: what is this site about. A blocking decision needs more. Who runs the domain, how long has it existed, and what is it doing right now? These six cases slip through category rules every day.
New domains look harmless
Scam and phishing domains are often used within days of registration. At that point they have no category yet, or a generic one such as Business or Shopping. A rule that allows those categories lets them through.
Parked domains change hands
A parked or for-sale domain shows ads or a sale page. Once it is sold, the buyer can point it at anything. The old category stays in your list while the content changes.
Lookalikes copy real content
A fake login page copies the colours, logo and words of the real brand. A content classifier may call it Online Banking or Email. Those are exactly the categories your policy allows.
Hacked sites keep their category
A real garden centre with a hacked folder still reads as Home & Garden. The phishing kit sits in a subfolder or on a subdomain. The category is right and the risk is still there.
Shared platforms host everyone
Site builders, blog hosts and file sharing services carry millions of honest pages and some abusive ones. Blocking the platform breaks real work. Allowing it without context lets abuse through.
Expired domains keep old trust
Domains that stop resolving, lapse or are registered again should not keep the trust they earned years ago. A category list has no way to show that change.
| Question | Web filtering category | Reputation score |
|---|---|---|
| What it answers | What is the site about? | How much can we trust this domain now? |
| How fast it changes | Slowly, when the content changes | Within days, when the state or threat picture changes |
| Main evidence | Page content and site purpose | Traffic, recent registrations, threat reports, hosting, name shape |
| Typical use | Acceptable use, compliance, reporting | Security blocks, warnings, risk scoring, triage |
| Example | Online Banking | 12, F, NEW_REGISTRATION|BRAND_LOOKALIKE |
Over-blocking without a score
- Teams block whole categories such as Uncategorized or Newly Registered to stay safe.
- Real small businesses, new projects and partner sites break with them.
- Help desks then add exceptions by hand, one ticket at a time.
Under-blocking without a score
- Lookalike and hacked sites inherit allowed categories.
- Parked domains that come back to life stay on old allow lists.
- Security teams learn about it from an incident, not from the filter.
How the score works
We combine 21 signals for each domain. Each one is weak on its own. Together they separate established sites from risky ones. Here we describe the signal types. The exact weights stay internal, so the score is hard to game.
The 21 signals behind every score
- Infrastructure footprintThe distinct jobs of the host names a domain runs
- Phishing historyPast phishing listings, fading with time
- Current phishing reportsThe domain itself is reported now
- Phishing host namesReported host names under the domain
- Malware reportsReports of malware on the domain
- Global popularityHow widely the domain is used worldwide
- Country popularityHow widely it is used in its own country
- Link authorityHow many other sites link to it
- Recent registrationRegistered or registered again in recent weeks
- Registration stateLapsed, or no longer registered
- Parked and for saleParking, ad or sale pages
- Expired and suspendedName servers of expiry and suspension services
- Hosting neighborhoodAbuse on the same server address and network block
- DNS operator riskShare of reported domains at its DNS operator
- Domain extension riskAbuse and parking rate of its extension
- Category contextEssential or abuse-prone category
- Email and verification recordsBusiness mail and service records in DNS
- Name shapeLength, hyphens, digits and random strings
- Brand lookalikes and typosBrand names next to login words, one-typo names
- Random subdomainsMany host names with no clear job
- Page checksReal site, placeholder, error or suspension notice, where we have checked the page
Popularity and real traffic
How widely the domain is used, measured by several independent traffic and link signals. When they agree, we trust the result more. A single signal on its own counts less.
Raises trustRecent registrations
Domains registered in recent weeks, and domains whose registration lapsed recently, get less trust until they build a record.
Lowers trustThreat reports
Phishing and malware reports for the domain itself and for host names under it. Reports against single sites on shared platforms count against those sites, not against the platform.
Lowers trust stronglyContent state
Parked, for-sale, expired and suspended domains, seen from their name servers. Where we have checked the page itself, also a real site, a placeholder, an error or a suspension notice. A parked page removes a lot.
Raises or lowersHosting neighborhood
How much known abuse sits on the same server address, the same network block and the same DNS operator. Abuse clusters. A domain in a bad neighborhood carries some of that risk.
Lowers trustExtension risk
How abuse and parking spread across top-level domains, measured over hundreds of millions of registered names. Some cheap extensions carry far more abuse. Restricted ones such as government and education raise trust.
Raises or lowersName shape
Length, digits, hyphens and random-looking strings. Brand names inside the domain, brand names next to words such as login or verify, and typos of brand names. Proven popular sites are not judged on their name.
Lowers trustBusiness records
Whether the domain has business email and service verification records. Real organisations leave these traces. Throwaway domains rarely do.
Raises trustInfrastructure footprint
The distinct jobs of the host names a domain runs: login, mail, API, CDN, download and more. More real roles mean more trust. Random host names earn nothing.
Raises trust stronglyPast threat reports
Phishing reports from our feed's history, even after the domain was cleaned. Recent, long and repeated listings weigh most. Old single reports fade.
Lowers trust, fades with timeCategory context and data depth
Essential categories such as government, education and banking add a little trust. Categories that attract abuse, such as anonymizers and file sharing, remove a little. When we have little evidence, the score stays near neutral and says so.
Raises or lowersHow the signals combine
The score is built in five steps. The same steps run for every domain, so two domains with the same evidence get the same score.
- Start neutral. Every domain starts in the middle of the scale. Unknown is not the same as bad, and not the same as good. Weak signals on their own, such as the extension or the shape of the name, can lower a score within the neutral band but never make a domain suspicious.
- Add evidence. Each signal adds or removes points. Strong, independent evidence moves the score more than weak or single-source evidence.
- Apply ceilings. Strong risk evidence sets a maximum. A domain on a phishing list, a parked domain or a domain registered a few days ago cannot reach a high grade, however clean the rest looks.
- Protect proven sites. Very popular domains get only small changes from name shape and neighborhood signals. On large sites those signals mostly produce false alarms.
- Explain. We keep up to four reason codes per domain, strongest first. Good scores show what earned the trust. Poor scores show what took it away.
Reputation is trust, not a content rating
An adult site or a betting site can be a long-running, well-run business. Its reputation can be good. Whether you allow it is a content decision. We keep content risk in a separate column of flags, so the two decisions never mix.
Five grades, each with a suggested action
The score runs from 0 to 100, higher is more trustworthy. Grades group the scores into bands that map to simple policy actions. The shares below come from the current database of 20M domains.
Trusted
Established, widely used, with real operations and no risk evidence.
Good
Real sites with normal signals and some traffic or business records.
Neutral
Little evidence either way. Many small or quiet sites sit here. Weak signals alone never push a domain below C.
Suspicious
Risk evidence such as a lookalike name, a dead site, a parked page, threat reports or a lapsed registration.
High risk
Threat reports, very new registrations, parked, for-sale or expired domains.
Grades are a starting point
Many teams block F everywhere, warn on D and use C together with the category. Sensitive groups, such as finance staff or students, can use a stricter line. You can also use the raw score if your system works with thresholds.
Every score comes with its reasons
Reason codes tell your analysts and your users why a domain got its grade. They are short, stable and easy to map to your own messages. These are the codes in the file.
| Code that raises trust | What it means |
|---|---|
| POPULAR_TOP1K | Among the most used sites on the web |
| POPULAR_TOP10K | Among the top ten thousand sites |
| POPULAR_TOP100K | Among the top hundred thousand sites |
| POPULAR_TOP1M | Among the top million sites |
| REAL_USER_TRAFFIC | Real visitors seen, below the top million |
| LINK_AUTHORITY | Many links from other sites point to it |
| RESTRICTED_TLD | Government, military or education extension |
| LOW_ABUSE_TLD | Extension with little abuse and parking |
| ESSENTIAL_CATEGORY | Government, education, banking, health or similar |
| BROAD_INFRASTRUCTURE | Runs four or more real infrastructure roles |
| INFRASTRUCTURE_ROLES | Runs one to three real infrastructure roles |
| BUSINESS_EMAIL_DNS | Business email and service verification records |
| EMAIL_DNS | Basic email records in DNS |
| LIVE_SITE | Our page check found a real site with readable content |
| RESOLVES | Registered and has an IP address in DNS |
| Code that lowers trust | What it means |
|---|---|
| PHISHING_LISTED | The domain itself is reported for phishing |
| PHISHING_HOSTS | Host names under the domain are reported for phishing |
| PHISHING_REPORTS | Reports on a popular domain, often a hacked page |
| MALWARE_LISTED | Reported for spreading malware |
| MALWARE_REPORTS | Malware reports on a popular domain |
| PAST_PHISHING_REPORTS | Reported for phishing in the past, fading with time |
| PHISHING_ON_SUBDOMAIN | Phishing pages on a subdomain of an established site |
| NEW_REGISTRATION | Newly registered or registered again in the last 30 days |
| RECENT_REGISTRATION | Newly registered or registered again one to two months ago |
| YOUNG_DOMAIN | Registered within the last few months |
| PARKED_OR_FOR_SALE | Shows a parking, ad or sale page |
| EXPIRED_OR_SUSPENDED | Points to an expiry or suspension service |
| REGISTRATION_LAPSED | Dropped out of registration recently |
| NOT_IN_REGISTER | The name is not registered |
| NOT_RESOLVING | The name does not resolve any more |
| NO_WEB_SERVER | Registered, but no working web address |
| SITE_ERROR | Server errors or missing pages |
| PLACEHOLDER_PAGE | Default server page, coming soon or under construction |
| SUSPENDED | Hosting account suspended |
| NO_READABLE_CONTENT | The site returns almost no text |
| SEIZED_OR_SINKHOLED | Seized by authorities or taken over by a security sinkhole |
| Name, hosting and context codes | What it means |
|---|---|
| BRAND_LOOKALIKE | Brand name with words such as login or verify, or on a risky extension |
| TYPOSQUAT | One typo or swapped character away from a brand |
| BRAND_NAME_USE | Contains a brand name it does not own |
| SUSPICIOUS_NAME | Login or account words with hyphens or a risky extension |
| RANDOM_LOOKING | Looks machine generated |
| RANDOM_SUBDOMAINS | Many random host names with no clear job |
| LONG_NAME | Unusually long name |
| MANY_HYPHENS | Two or more hyphens |
| MANY_DIGITS | Many digits in the name |
| BAD_HOSTING_NEIGHBORHOOD | Shares servers with many reported domains |
| RISKY_DNS_PROVIDER | DNS operator with a high share of reported domains |
| RISKY_TLD | Extension with a high share of abuse or parking |
| HIGH_ABUSE_CATEGORY | Category that attracts abuse, such as anonymizers |
| NO_TRAFFIC_SIGNAL | No sign of real visitors yet |
| LIMITED_DATA | Little evidence, score stays near neutral |
| Content flag (separate column) | What it means |
|---|---|
| ADULT | Adult content |
| GAMBLING | Betting, casino or lottery |
| DATING | Dating and personals |
| ALCOHOL TOBACCO | Alcohol or tobacco sales and promotion |
| WEAPONS VIOLENCE | Weapons or violent content |
| DRUGS | Illegal drugs |
| HATE | Hate or discrimination |
| LINGERIE | Lingerie and swimwear |
| ANONYMIZER | VPN, proxy or anonymizer, can bypass filters |
| P2P | File sharing and peer to peer |
| CRYPTO | Cryptocurrency services |
| USER_CONTENT | Shared platform: content made by many users, judge single hosts |
| PAST_PHISHING | Reported for phishing in the past |
| PHISHING_ON_SUBDOMAIN | Phishing pages were found on a subdomain |
What the rows look like
Five columns, one row per domain, the same domain key as the category file. Below are real rows from the file with the names masked. The sample download has 1,000 rows across all grades with the web filtering category added.
One key, one join
The domain column matches the category file row for row. Load both into your lookup table and every decision can see the category, the grade and the reasons at once.
No live lookups needed
The scores ship in the file. Your product can decide offline, inside a firewall or a resolver, without sending user traffic to anyone.
Comes with updates
Reputation changes faster than categories. When you take database updates, you receive current scores with them, so expired, parked and newly reported domains move down.
Where teams use the reputation score
The score is built for products that make a decision per domain, many times a second, and need a short reason they can show.
Firewalls and UTM
Add a reputation rule above the category rules. Block F in every category. Send D to a warning page or to inspection. Keep A and B on the fast path, which cuts load on deep inspection.
DNS filters
Answer F domains with a block page at the resolver. Log D domains for review. Use the reason codes in the block page, so users and help desks see why without opening a ticket.
Browser security
Show a warning before a user types a password on a D or F domain. Treat BRAND_LOOKALIKE and TYPOSQUAT with extra care on login forms. Leave proven sites alone to avoid warning fatigue.
Fraud and payments
Score the website of a new merchant at onboarding. Score the email domain of a new sign-up. A brand-new domain with no traffic and a lookalike name deserves a second look before money moves.
Email and link checks
Rate every link in a message by its domain. Rewrite or hold messages whose links point to F domains. Use NEW_REGISTRATION as a strong hint in invoice and password reset messages.
SOC triage
Sort alerts by the reputation of the domains involved. Alerts that touch A domains are often noise. Alerts that touch new, parked or reported domains go to the top of the queue.
A simple policy that combines both
| Grade | Category your policy allows | Category your policy blocks |
|---|---|---|
| A and B | Allow | Block by category |
| C | Allow and log | Block by category |
| D | Warn, isolate or block for sensitive groups | Block |
| F | Block | Block |
Content flags stay with the category side of the table. A domain flagged GAMBLING with grade A is blocked by your gambling rule, not by the score. A domain in Business with grade F is blocked by the score, even though Business is allowed.
How we test that the score makes sense
A trust score is only useful when its extremes are right. Every time we build the scores, we check them against groups of domains whose answer we already know.
Infrastructure separates real from throwaway
Two in three of the thousand most used sites run four or more real infrastructure roles. Among reported phishing domains and new registrations it is fewer than one in two hundred.
Past phishing keeps counting
Domains reported for phishing in the last two months land in grade F in 85% of cases, even after removal. Very popular sites keep their grade and get a flag.
The most used sites score high
Every domain in the database that ranks among the thousand most used sites gets grade A. Popular sites with reports keep a visible reason code.
Random domains, checked by hand
We draw random domains from every grade and look at each live site by hand, without seeing its grade. In the latest check, 96 of 100 domains in grade F were parked, for sale, dead, empty or abusive, and 2 of 160 domains in grades A and B were risky sites.
Parked, for-sale and new domains score low
Domains on parking or sale services and domains registered in the last 30 days have a ceiling, so they land in grade D or F. About two in three parked domains and three in four new registrations land in F.
Content is not punished
Adult and gambling domains score like other sites with the same traffic and signals. Their risk shows in the content flags, not in the trust score.
Category and reputation in one licence
Domain reputation scores are included with the 20M Web Filtering Database. Tell us what you build and how many users it serves, and we will quote a licence that fits.
Tell us how you will use the scores
We size the licence to your product and your users. A short note helps us answer in one message.
- What the data goes into: firewall, DNS filter, browser tool, fraud system or your own network.
- Roughly how many users or devices it protects.
- Whether you need the category, the reputation score or both.
Prefer to look first? Download the 1,000-row sample or read about the Web Filtering Database and its 58 categories.
Message Sent!
Thank you. We will get back to you within one business day.
Questions about the reputation score
Does the reputation score replace web filtering categories?
No. The category says what a site is about. The reputation score says how much you can trust the domain right now. Most policies use both: the category for acceptable use and the score for security decisions.
Why can an adult or gambling site have a good reputation?
Reputation measures trust, not content. A long-running, widely used site can be trustworthy even if your policy blocks its content. We keep content risk in a separate column of flags, so you can block the content and still see that the domain itself is not a threat.
What does a C grade mean?
C means neutral. We found no strong risk evidence and not much positive evidence either. Many small, quiet but legitimate sites sit here. Weak signals alone, such as the extension or the shape of the name, never move a domain below C. Treat C domains with your category policy and log them.
Can a domain's score change?
Yes. A domain that expires, gets parked, is registered again or appears in threat reports will drop. A new domain that builds real traffic will rise. When you take database updates, the scores come with them.
How do you treat sites on shared platforms such as site builders and blog hosts?
Reports against single sites on a shared platform count against those host names, not against the platform itself. The platform row carries the USER_CONTENT flag, so you know that content on it is created by many different users.
Do you publish the weights of the score?
No. We describe the signal types, the grades and every reason code. The exact weights stay internal, which makes the score harder to game for people who register domains to abuse them.
Why do infrastructure roles raise trust?
A real organisation runs hosts for login, mail, APIs, CDN traffic and downloads. Building and running them costs money and time, so throwaway and phishing domains almost never have them. More distinct roles add more trust; one host or random names add nothing.
Does a domain keep a lower score after a phishing page was cleaned?
Yes, for a while. A past phishing report keeps lowering the score after removal. Recent, long or repeated listings weigh most, and the penalty fades over the following months. Very popular sites get a flag instead of a low grade.
How do I get the reputation scores?
Domain reputation scores are included with the 20M Web Filtering Database. Ask us for a quote and tell us how you will use the data, and we will size the licence to your product or network. Prices for the database sizes are on our pricing page.