← all guides

IP reputation feeds and how an exit gets flagged

The exit that was flagged before you touched it

You rotate to a fresh exit, send one polite request, and get a challenge page. Nothing about your request was unusual. It was the first thing that address had done for you. But the address had a past, and the site you asked already knew about it.

That’s IP reputation at work. Somewhere, one or more systems have a record of what traffic from that address, or from its neighbours, did before. It might have sent spam. It might have scanned servers for open ports. It might have been part of a pool someone used to hammer a login page last week. The target, or the vendor protecting it, consulted that record and decided to be careful.

This piece covers where those records come from, how an exit ends up carrying a reputation it didn’t earn under your name, and how to check an exit’s history yourself before you trust it with a job. It’s about history. The companion piece on whois and reverse DNS on an exit node covers identity: who owns the address and what its hostname says. Those are related, since a hosting range is identified by its owner and judged by its history, but they answer different questions.

The piece on what a shared proxy pool really shares touches on reputation as one of the things a pool hands between customers. Here the focus is the mechanics: the feeds, how scores accumulate, how long they last, and what you can see from outside.

None of this is a guide to finding addresses that slip past reputation checks. Reputation is a site deciding how much benefit of the doubt to give, based on evidence. The point of understanding it is to buy and use exits sensibly, and to avoid adding to a bad record yourself.

Reputation is history attached to an address

An IP reputation score is a summary of evidence about past behaviour associated with an address. The evidence comes from many sources, it’s collected and combined by different organisations for different purposes, and there’s no single authoritative score. There are many, and they disagree.

Three levels matter. The individual address carries its own history. The range it sits in, often the /24, carries the history of its neighbours; the piece on subnet diversity explains why systems group addresses that way. And the network that announces it, the autonomous system, carries a reputation too, which is why some hosting providers are treated more warily than others.

Reputation also has a type. An address might be flagged for sending email spam, for scanning, for credential stuffing, for being a known proxy or VPN exit, for being a Tor exit, for hosting malware, or for generating abusive web traffic. A site deciding whether to trust a request cares about some of those types far more than others. A mail server cares about spam history. A login page cares about credential stuffing. A retailer’s bot management cares about scraping and proxy use.

And reputation has an age. Evidence from yesterday counts for more than evidence from a year ago. Most systems decay old evidence, though at very different rates.

So when people say an address has “bad reputation”, the useful follow-up questions are: bad according to whom, for what, at which level, and how recently. The answers decide whether it matters for your job.

Spam and abuse blocklists

The oldest reputation systems are DNS-based blocklists, built originally to protect mail servers. A mail server receiving a connection looks up the connecting address in one or more lists, and rejects or scores the message if it’s listed. The lookup works over DNS: the address’s octets are reversed and queried as a name under the list’s zone, and a listed address returns an answer.

Different lists track different things. Some list addresses caught sending spam. Some list addresses of compromised machines, often home computers infected with malware that sends spam or joins botnets. Some are policy lists that name address ranges which shouldn’t be sending mail directly at all, such as dynamic residential ranges, without implying anyone there did anything wrong. Some list ranges controlled by known spam operations.

Web-focused systems borrow from these lists. A bot management vendor or a site’s own firewall may treat a listing for compromised hosts as a signal of risk on web traffic too, on the reasonable theory that an infected machine is more likely to be sending automated requests. Policy listings are weaker evidence for web purposes, since they describe the kind of range rather than any behaviour.

For proxies, these lists matter in two ways. Residential proxy exits sit in consumer ranges, which appear on policy lists by design, and some of those devices may be on compromised-host lists for reasons that have nothing to do with the proxy network. Datacenter exits appear on spam lists when someone on the same range, or the same address earlier, sent spam.

Many of these lists have usage terms. Some are free for low-volume or non-commercial queries and require a subscription beyond that. Some refuse queries arriving through large public DNS resolvers. Read the terms before building lookups into anything automated.

Community abuse reports

A second family of sources is abuse reporting databases, where server operators report addresses that attacked or abused them. A server administrator sees repeated failed logins from an address, or a burst of requests probing for vulnerable software, and submits a report with a category and a timestamp. Other operators can then look the address up and see how many reports it has, how recent they are, and what kinds of abuse were reported.

These databases are useful because they’re broad and current. They’re also noisy. Reports come from anyone, categories are applied inconsistently, and an address that’s shared among many users, like a mobile carrier’s NAT address or a residential proxy exit, can accumulate reports that describe some of the people behind it rather than all of them.

Many of these services have a free, rate-limited lookup and a paid tier for higher volumes. They typically present a confidence score based on the number of distinct reporters and the recency of reports, alongside the raw reports.

For evaluating a proxy exit, the useful signals are the number of distinct reporters, the categories, and the dates. A single old report is weak evidence. Many recent reports from different reporters, in categories like web attacks or brute force, suggest the address has recently been used for exactly the kind of traffic sites defend against, and your requests will be judged in that light.

Reports on a shared carrier address need more care, since the carrier’s own customers generate traffic too. The piece on carrier NAT and who else sits on your exit covers why a mobile address is always a crowd.

Sensor networks and threat intelligence feeds

A third source is sensor networks: systems deliberately placed to observe unsolicited traffic. Security companies and research projects run large numbers of servers and address ranges that have no legitimate reason to receive traffic. Anything that connects to them is, almost by definition, scanning, probing or attacking. The addresses that do so get recorded and published as threat intelligence feeds.

Web honeypots work similarly on the application layer. Pages and paths that no legitimate visitor would request, including the hidden links and trap URLs described in the piece on honeypot links, identify clients that crawl indiscriminately. Some of that data stays with the site, and some ends up with vendors or feeds.

Threat intelligence feeds are commonly consumed by firewalls, security products and bot management systems. Some are published openly, some are commercial, and there are public projects that aggregate many open feeds into combined lists by category.

These feeds matter for proxies because the traffic they observe is often exactly what proxy networks get used for by their least careful customers: scanning, credential stuffing, aggressive crawling. An exit that some other customer used for a scan last week may appear on a scanner feed today. When you rent that exit for a polite job, the feed doesn’t know or care that the tenant changed.

The feeds also tend to be quick to add and slower to remove. A sensor sees a scan and records it immediately. Whether the address is removed after a quiet period depends on the feed’s policy.

Classification as a proxy or VPN

A fourth kind of reputation is classification rather than behaviour. Commercial IP intelligence providers maintain databases that label addresses by type: residential, mobile, business, hosting, and also flags such as known VPN exit, known proxy, known Tor exit. Fraud prevention, advertising verification and bot management systems buy this data and use it as one input.

These labels are built in various ways. Hosting ranges are identified from network registration and ownership. VPN and Tor exits are identified from published lists and by observing the services themselves. Proxy exits, including residential proxy exits, are identified by observing traffic patterns, by signing up for proxy services and recording the exits they hand out, and by correlating signals across many customer sites.

That last method matters for anyone buying residential or mobile exits. An address can be a genuine home or mobile connection, registered to a consumer ISP, and still carry a “proxy” flag in a commercial database because it has been observed acting as a proxy exit. From the target’s side, the registration says residential and the classification says residential proxy. Both are true.

Classification isn’t a verdict of wrongdoing. Plenty of legitimate traffic comes through VPNs and proxies. But many sites apply stricter rules to classified proxy traffic, especially in areas like account creation, login, checkout and anything involving promotions. A proxy flag on an exit is information about how the target may treat it, whatever your job’s intentions.

You can check how some of these databases classify an address, since many providers let you run a handful of free lookups. Different databases will disagree, sometimes sharply.

The reputation you can’t look up

The most important reputation for your job is usually one you have no direct way to see: the target’s own, and its vendor’s.

Large sites keep their own records of addresses, ranges, sessions and devices that behaved badly on their properties. Bot management vendors keep records across all the sites they protect, and a pattern seen on one customer’s site can raise scores on another’s. CDNs that also sell security products see a large share of web traffic and build reputation from it. None of this is published, and it’s often the deciding factor.

That’s why external lookups are only a partial check. An address can be clean on every public list and still be treated warily by a target that saw it misbehave last month, or by a vendor that saw the same pool abuse another customer. Equally, an address with an old spam listing may pass through a target’s web defences without any trouble, because that target doesn’t weigh spam history at all.

The only reliable test of how a target treats an exit is to send it a small, polite request and look at the response carefully. The piece on soft blocks that return 200 explains why the status code alone isn’t enough. Compare the content against what an ordinary connection receives. If they match, the target’s view of that exit is acceptable for now. If they don’t, the target has a view you couldn’t have looked up.

How an exit inherits a bad name

Understanding where reputation comes from makes it clear how an exit you’ve just been given can already carry one.

Datacenter addresses get reassigned. When a hosting customer leaves, their addresses go back to the provider and eventually to someone else. If the previous tenant ran a spam operation or a scanner, the address may still be listed when the next tenant, possibly a proxy provider, starts using it.

Shared pools pass reputation between customers constantly. If a provider sells access to the same exits to many customers, every customer’s traffic writes to the same record. One customer’s credential-stuffing run and another customer’s polite price check go into the same history. The piece on shared pools calls this the half that isn’t on anybody’s pricing page.

Residential exits are real devices with real owners. The same device may have been used by its owner for ordinary browsing, may be infected with unrelated malware, and may be serving as an exit for several proxy networks at once. Its reputation reflects all of that.

Mobile exits sit behind carrier NAT, so each public address carries the combined history of everyone behind it. That’s partly why sites are cautious about acting on mobile reputation, since blocking one address may block many real customers, but it also means a mobile exit’s record is never only yours.

And ranges spill over. An address with no history of its own inherits some suspicion from a range where many neighbours are listed.

How long a bad name lasts

Reputation decays, but not evenly, and the differences matter when deciding whether a flagged exit is worth keeping.

Spam blocklists usually have explicit policies. Some list addresses for a fixed period after the last observed spam and remove them automatically if nothing further happens. Some require the owner to request removal after fixing the problem. Policy lists don’t decay, since they describe the kind of range.

Abuse report databases typically show report history with dates, and their confidence scores give more weight to recent reports. Old reports remain visible, but they count for less.

Threat intelligence feeds vary widely. Some age out addresses after a quiet period of days or weeks. Some keep them much longer.

Commercial classification tends to be sticky. An address observed acting as a proxy exit may keep that label for a long time, because the provider has no easy way to confirm it stopped.

Private reputation inside targets and vendors is the least predictable. It may decay in days, or it may attach to a range or a pattern that keeps getting reinforced by new traffic.

The practical upshot is that “wait and it’ll clear” is sometimes true and sometimes not. An exit with a single recent spam listing may be clean in a week. An exit with a proxy classification may stay classified indefinitely, and that’s a property of what it is, not something to wait out.

Looking up an exit yourself

Here’s a sequence you can run on any exit, during a trial or before trusting an address with a job.

Start by confirming what the address is. Log the exit you actually get from an echo endpoint, since it may differ from the address on a list. Look up the network that announces it and the organisation that owns it. That tells you whether it’s hosting, consumer broadband or mobile, which frames everything else. The whois and reverse DNS piece covers this step.

Check a few DNS blocklists, respecting their usage terms and querying through a resolver they accept. Note which lists include the address and what each list tracks. A policy listing on a residential address is expected. A compromised-host listing is worth noting. A spam listing on a datacenter address says something about the range’s history.

Check an abuse reporting database. Look at the number of distinct reporters, the categories, and the dates of the most recent reports.

Check one or two IP intelligence lookups for classification: hosting, proxy, VPN flags. Expect disagreement between them.

Then run the only test that reflects your target. Send a small number of polite requests to the target through that exit and compare the content with what an ordinary connection gets.

Record all of it with dates. Reputation changes, and a record from a month ago is a starting point for the next check, not a conclusion.

If you want to automate this across a pool, do it gently. Check a sample of exits rather than every address every hour, cache results for a sensible period, and stay within each service’s free or paid limits. Hammering a reputation service with lookups is its own small abuse, and several of them will block you for it. A daily check of the exits you actually used, plus a weekly sample of the wider pool, is usually enough to spot a range going bad. Keep the target tests separate and even lighter: a few requests per exit, spaced out, only against targets the job already covers.

Reading the results sensibly

The results from those lookups need interpretation, and it’s easy to overreact or underreact.

Separate policy from behaviour. A residential address on a policy list is listed because of the kind of range it’s in. That tells you nothing about what happened on that address. A listing for observed spam, scanning or attacks is behavioural and more relevant.

Match the type to the target. A spam listing matters a lot if your job involves sending mail, which a scraping job shouldn’t, and less for most web targets. A web-attack or brute-force report matters more for web targets. A proxy classification matters most on targets that treat proxies strictly, often exactly the targets where you should be asking whether scraping is appropriate at all.

Weigh recency and breadth. Many recent reports from many reporters are strong evidence. One old report is weak.

Look at the range as well as the address. If a whole /24 of datacenter exits shows spam and scanning history, the provider’s range has a problem that will keep resurfacing as addresses rotate.

And give most weight to the target test. If the target serves the exit the same content as an ordinary connection, public listings are less important for this job. If it doesn’t, the reason may be something you can’t see, and the right response is the same regardless of which list, if any, carries the address.

Three exits, looked up by hand

A made-up example shows how the same lookups can point in different directions. None of these addresses or results is real.

Exit A is a datacenter address on a hosting provider’s network. It appears on one spam blocklist, listed for sending spam about a month ago. The abuse database shows two reports, both older than that. It’s classified as hosting by every IP intelligence lookup, and one flags it as a known proxy. Through it, the target, a public documentation site, returns the same content an ordinary connection gets. For this job, Exit A is fine. The spam listing reflects a previous tenant’s mail activity, which this target doesn’t weigh, and the target test is clean.

Exit B is a residential address on a consumer ISP. It’s on a policy blocklist, as consumer ranges usually are, and nowhere else. The abuse database has nothing. One IP intelligence lookup flags it as a residential proxy exit; another shows it as ordinary residential. Through it, the target, a retailer, returns a product page with the price block missing and a challenge script in the page. For this target, Exit B is a problem, and none of the public lookups would have told you why with certainty. The proxy flag is a plausible explanation. The retailer’s own records are another.

Exit C is a mobile carrier address. It has a scattering of abuse reports from different reporters over the past few weeks, in mixed categories. It isn’t on any spam list. Classification says mobile. The target returns normal content. The reports most likely describe other users behind the same carrier NAT address; for this target, they don’t matter today. But the address is shared, so the record will keep changing, and the check is worth repeating.

The lesson across all three is the same one. Public lookups tell you what kinds of history an exit carries. The target test tells you whether that history matters for the job in front of you.

What to do with a flagged exit

Once you know an exit carries a history that matters for your target, the options are simple.

Drop it for that target. Most pools let you exclude addresses or ranges, or at least rotate away from them. If you manage your own pool, keep a per-target exclusion list with the reason and date, and recheck periodically.

Tell the provider. A good provider wants to know which exits are flagged, and may pull them, rotate the range, or request delisting where the address belongs to them. Their response is also a useful signal about how they run the pool.

Don’t treat rotation as a cleaning process. Cycling through exits until one passes a target’s check, after the target has flagged several, runs straight into the behavioural scoring covered in the piece on signals anti-bot vendors score, and it adds your traffic to the record of every exit it touches.

Don’t request delisting for addresses you don’t control. Delisting processes expect the requester to be the address owner who has fixed the cause. For a rented exit, that’s the provider.

And consider the job itself. If most exits in a pool are flagged by your target, the pool may be the wrong product, or the target may simply not want automated traffic. The second possibility deserves an honest look before the first.

Not becoming the previous tenant

Every piece of reputation described here was written by somebody’s traffic. On a shared pool, a rented range or a residential network, the next customer inherits what you do.

That’s a practical reason, on top of the obvious one, to keep a scraping job’s traffic compliant. Traffic that stays within a target’s robots rules and rate limits, identifies itself where possible, and backs off when challenged is unlikely to generate abuse reports, trip honeypots or feed a vendor’s model with a pattern worth flagging. It leaves the exit’s record roughly where it found it.

Traffic that hammers a target, ignores robots or walks into traps does the opposite. It generates reports against the exit and its range, and those reports outlast your use of the address. The provider’s other customers pay for it, and so, eventually, do you, because the pool you rely on gets worse.

If you run infrastructure, make it easy for people to report abuse from your addresses. A working abuse contact in the registration record, answered promptly, lets site operators tell you about a problem instead of listing it. It’s also part of what separates a provider that cares about its ranges from one that doesn’t.

What a provider should be doing about it

A buyer can check exits one at a time. A provider can see the whole pool, and the good ones do the work that buyers can’t.

That starts with monitoring. A provider running its own ranges can check them against the main blocklists and abuse databases continuously, and see listings appear as they happen. When a range starts collecting spam listings or abuse reports, that’s usually one customer’s traffic, and the provider can find and deal with it before the whole range is marked.

It continues with acceptable use enforcement. The piece on what a proxy provider’s acceptable use policy actually forbids covers what those policies say. Reputation is where you see whether they’re enforced. A provider whose ranges stay reasonably clean over time is almost certainly removing customers who generate abuse. One whose ranges are listed everywhere is probably not.

It includes delisting and rotation. Where the provider owns the addresses and has fixed the cause, it can request removal from lists that support it. Where a range is too damaged, it can retire it.

And it includes honesty with customers. A provider should be able to say, in general terms, how it monitors reputation, how it handles abuse reports against its addresses, and whether it will let you exclude specific exits or ranges from your traffic. Those are fair questions to ask during a sales conversation, and vague answers are informative.

For residential and mobile pools, providers have less control, since the underlying addresses belong to ISPs and carriers. They can still monitor, still enforce acceptable use against their own customers, and still exclude exits that are clearly damaged.

What I can’t tell you

I can’t tell you which reputation sources a specific target uses, how it weights them, or how long its own records last. That’s private, it varies by site and vendor, and it changes.

I can’t tell you that an exit clean on public lists will be treated well by your target. Public lookups show part of the picture. The target test shows what matters.

I can’t give you accuracy figures for any list, database or classification service. They measure different things with different methods, and they disagree with each other routinely.

What I can say is that reputation is history, collected by many parties for different reasons, and an exit carries all of it into your job. Look up what you can, weigh it by type and age, test against the target itself, drop what doesn’t fit, and run traffic that won’t make the next tenant’s lookup worse.

If you want the related pieces on whois and reverse DNS, shared pools and free proxy lists, Proxy Scraping has the full set of guides from the operator side.

Get new guides and videos first — join the Telegram channel.

proxies
Need proxies that survive the block wall?

Singapore Mobile Proxy runs real 4G/5G mobile IPs on rotating SIMs — the carrier-grade addresses most of these targets still trust.

see plans →
read on
More scraping guides

The rest of the field manual: target-site playbooks, library walkthroughs, provider reviews, and anti-bot troubleshooting.

browse all guides →