Link spam policy
A link scheme is any link arrangement that exists mainly to move rankings rather than to help a reader. That is the short version, and it is worth one correction before you read anything else about it: Google mostly stopped using the phrase. Its spam policies call the section link spam now, and define it as creating links to or from a site primarily for the purpose of manipulating search rankings.
The old name survives in one place that matters, which is Search Console's own manual actions documentation, still describing sites that participate in link schemes. So the term is not wrong. It is just older than the policy everyone is arguing about, and reading the current wording settles most of those arguments in about four minutes.
The short answer
Quick reference
Nine situations, one question, and this table is the whole argument. Everything below it is the reasoning and the caveats.
A writer cites your page because it was the best source for their point
An independent editorial decision. No bullet in the policy touches this.
A genuine business partnership both sides describe publicly
Integration partners, suppliers, association members, event sponsors.
A sponsored placement carrying rel="sponsored"
Named as acceptable in the policy itself, not a loophole.
An occasional reciprocal link between two relevant sites
A guest article you wrote, published on merit, with a natural anchor
Fine on its own. The bullet is about scale and optimized anchors, so volume is what changes this.
Systematic link exchanges run to a monthly target
The excessive link exchanges bullet, almost word for word.
A paid link that passes ranking credit with no attribute
The first bullet in the policy, and the least ambiguous one.
Links from a network of sites you own or control
Not named by acronym, and comprehensively covered by the bullets it triggers.
Automated or software-generated link creation
Named directly, with no qualifier about scale attached to it.
Risk here means how likely a pattern is to be devalued or actioned, read against Google's published policy. It is our editorial judgment, not a score Google publishes, and we would rather say that plainly than imply we have numbers nobody outside Google has.
Contents
Written for somebody who has to decide whether a specific link is worth taking, or explain to a client why last year's link building stopped working. Jump to whichever question you arrived with.
Definition
A link scheme is any arrangement where links are created primarily to influence search rankings rather than to point a reader at something worth reading. It covers buying, selling, trading, manufacturing and automating links, and the common thread through all of it is purpose rather than method.
01
Link spam
Google's current term for the same territory. The policy defines it as creating links to or from a site primarily for the purpose of manipulating search rankings, then gives examples. The definition is about why, not what.
02
Backlink scheme
An informal way of saying the same thing with the emphasis on inbound links. Not a term Google uses, which is worth remembering when somebody quotes a rule about it.
03
Unnatural link
The phrase Search Console uses when it issues a manual action. It describes links a reviewer judged artificial, deceptive or manipulative, and it exists in an inbound and an outbound version.
04
Manipulative link
Any link whose reason to exist is the ranking signal. The most useful working definition on this page, because it is the one you can apply to a single email in front of you.
05
Toxic backlink
A vendor metric. No search engine publishes it, no Google documentation uses it, and treating a toxicity score as a policy verdict is how clean profiles end up disavowed for no reason.
Terminology
For years the reference document was a Webmaster Guidelines page titled Link schemes. When Google rewrote its guidelines into the current spam policies, that page went and the section that replaced it is called link spam. Same subject, different framing, and the framing is the point.
The policy is not a list of banned tactics. It is one sentence about intent, with eleven examples attached.
A scheme sounds like a thing you join. Spam sounds like a thing you produce. The second is closer to how the policy actually reads, because almost every bullet in it describes a behavior at volume rather than a club with members. That is also why the tactic-by-tactic arguments people have about this never conclude. The policy is not a list of banned tactics. It is one sentence about intent with examples attached.
The old phrase does survive in one place worth knowing about, and it is not a trivial one. Search Console's manual actions documentation still describes buying links or participating in link schemes as a violation, in the text of the action a real reviewer applies to a real site. So if you are searching for the term because somebody used it in a meeting, they are not out of date. They are quoting the enforcement half rather than the policy half.
If your question is narrower than this page, we have written the exchange-specific version of it: where Google's policy draws the line on link exchanges covers the trading half in detail.
The policy text
Quoted in full
Link spam is the practice of creating links to or from a site primarily for the purpose of manipulating search rankings.
One sentence, and three parts of it are doing work. Held against it, the two exchange structures compared come out closer than the pitch for either suggests. To or from, which means this is not only about what points at you. Primarily, which concedes that links can have more than one reason to exist. And manipulating search rankings, which is a statement about purpose that no audit tool can measure for you.
That last part is why there is no threshold anywhere in this policy, and why every article promising you a safe percentage invented it. Purpose is not a number. The examples that follow the definition are Google showing its working, not a checklist you can clear.
Which leaves you doing the thing nobody wants to do, which is making a judgment about your own motives. The good news is that the judgment is usually easy. People know why they bought the link.
The examples
Worth reading in full at least once, because the list is more specific than the summaries of it. Several bullets catch practices that most teams would not think of as link building at all.
| What the policy lists | What it actually covers | Where teams get caught |
|---|---|---|
| Buying or selling links for ranking purposes | Money for links or for posts containing links, goods or services for links, and sending someone a product in exchange for a write-up with a link | Gifted product with an unqualified link is the one people forget. The policy names it explicitly. |
| Excessive link exchanges | The link to me and I will link to you arrangement, plus partner pages built exclusively for cross-linking | Nobody counts their own exchanges, so nobody knows whether they are near excessive. |
| Automated link creation | Using programs or services to create links to your site | Bought from a marketplace, delivered by software, and no qualifier about scale in the bullet at all. |
| Links required by contract | Requiring a link in a Terms of Service or agreement without letting the other party qualify it | Agency footers, theme licenses, plugin terms. Usually legacy, usually nobody's decision. |
| Text ads that do not block ranking credit | Paid text links and text advertisements passing PageRank | The sidebar ad unit somebody hardcoded in 2019 with a plain anchor tag. |
| Advertorials and native advertising | Paid articles whose links pass credit, and optimized anchors in articles, guest posts or press releases distributed on other sites | Press release syndication with commercial anchors, still sold as PR. |
| Low quality directory or bookmark links | Directory and bookmark submissions with no editorial standard | Inherited from an older campaign, and mostly harmless clutter unless you built them. |
| Keyword-rich links in widgets | Hidden or low quality links embedded in widgets distributed across sites | Badges, calculators and embeddable tools that carry an anchor the host never chose. |
| Widely distributed footer and template links | The same link rendered across the footers or templates of many sites | Built by an agency, a network, or a white label supplier three vendors ago. |
| Optimized forum comments and signatures | Commercial anchors in forum posts and profile signatures | Cheap, old, and still being sold as community engagement. |
| Low-value content built to move links | Creating thin content primarily to manipulate linking and ranking signals | The catch-all bullet, and the one a scaled content program is most likely to trip. |
Summarized from Google's link spam policy. Read the original: it is one screen long and the wording carries the argument better than any paraphrase, including this one.
The mechanism
Links work as a ranking signal because they are supposed to be expensive. Not in money, in judgment. Somebody with their own reputation decided your page was worth pointing their readers at, and that decision is hard to fake at volume by anyone who has not done the work.
Manufacture the endorsement and you have not gained a signal. You have degraded one.
Manufacture the decision and you have not gained a signal, you have degraded one. If any company could produce a thousand endorsements by wiring money to a list of domains, the endorsement would stop meaning anything, and the whole mechanism that made links useful in the first place would quietly stop working for everyone.
That is the entire logic of the policy, and it is worth holding on to, because it predicts things a rule list cannot. Whatever the next tactic is called, if it works by manufacturing the appearance of independent judgment, it is in scope. Nobody at Google has to write a bullet about it first.
The value itself moves along links, which is the part the manipulation is aimed at. We have written about how that flow works on your own site in internal links versus backlinks, and the short version applies here too: external links raise what a site has, and everything else is distribution.
Types, part one
The shapes where something changes hands, drawn from the same set as the 25 backlink types. These are the ones with the clearest policy language attached, and also the ones with the most legitimate versions sitting next to them.
Would this link still exist if links counted for nothing?
01
The most straightforward violation in the policy, and the easiest to fix. Buying a link is not the problem. Buying a link that passes ranking signals without a rel attribute is. The money is allowed to exist as long as the ranking claim does not travel with it.
The tell: An invoice, a live dofollow link, and no sponsored or nofollow attribute anywhere on it.
02
Two sites agreeing to link to each other. Ordinary once, a program at ninety, and the policy language is about the second. What makes exchanges genuinely hard to govern is that each individual trade looks reasonable in isolation, and almost nobody has a number for how many they are running.
The tell: A spreadsheet of partner domains sorted by domain rating, with a monthly target above it.
03
A links to B, B links to C, C links back to A. Sold as the safe version because it removes the visible pair. It removes exactly one signal and leaves the anchors, the placement, the timing and the relevance untouched, which are the signals that usually give the arrangement away.
The tell: Somebody proposes the third hop specifically so the trade will not be visible.
04
Paid editorial content, and press releases distributed with commercial anchors. The policy names both, and it names optimized anchor text in articles, guest posts and press releases as its own case. This is the bullet most PR retainers are quietly failing.
The tell: The anchor in the release is a product keyword rather than the company name.
05
The distinction is whether anyone is refused entry. A real industry directory or association listing has a standard, an audience and a reason to exist. A directory that accepts anything for a fee has one customer, and it is not the reader.
The tell: Approval is automatic, the categories are enormous, and the only visitors are other link builders.
Types, part two
Where the other side of the link is not an independent publisher at all, because you built it, bought it, or a script created it. These carry the most risk, and the least ambiguity.
01
Sites, usually on expired domains with leftover authority, whose reason to exist is linking to a target. The policy never uses the acronym, which occasionally gets read as a gap. It is not one: the arrangement fails several bullets at once, and there is no version of a PBN where an independent editor made a decision.
The tell: You control both ends of the link and would not describe the arrangement to a client.
02
Pages whose function is holding outbound links, with no audience of their own. The tell is a ratio rather than a judgment call: many outbound links, no traffic, no reason for a human to be on the page.
The tell: The page ranks for nothing and links to sixty unrelated domains.
03
Software or a service generating links at a rate no person could review. Named directly in the policy with no qualifier about volume, which makes it one of the few bullets that does not depend on scale. The lack of editorial review is the violation, not the count.
The tell: The links appeared faster than anyone could have read the pages they sit on.
04
Not guest posting, which is a legitimate thing writers do. Networks: a pool of sites that publish anything, sold by the placement, where the article exists to carry the anchor. The giveaway is that the writer, the topic and the publication have no relationship to each other.
The tell: You are choosing the anchor text and the publication does not ask about the article.
05
An embeddable tool, badge or theme that carries a keyword-rich link back, installed by people who never chose the anchor. The policy names this specifically. A widget that links with your brand name, disclosed, is a different thing from one that hides a commercial anchor in the markup.
The tell: The embed code contains an anchor the host would not have written.
06
Widely distributed links in footers and templates, named in the policy. One agency credit in the footer of a client site is normal practice. The same commercial anchor rendered across thousands of pages on dozens of sites is the thing the bullet describes.
The tell: One agreement produced ten thousand links and one referring domain.
The exception people miss
Quoted in full
Google does understand that buying and selling links is a normal part of the economy of the web for advertising and sponsorship purposes. It's not a violation of our policies to have such links as long as they are qualified with a rel="nofollow" or rel="sponsored" attribute value to the <a> tag.
That paragraph sits at the end of the link spam section and it contradicts a large amount of what gets written about paid links. Sponsorship is not banned. Advertising is not banned. Paying a publisher is not, on its own, a policy problem at all.
What is banned is the ranking credit traveling with the payment. Add rel="sponsored", which is one of three real qualifying values, and the placement is compliant, the traffic still arrives, the brand exposure still happens, and the only thing you gave up is the part you were not entitled to.
Which reframes most sponsorship arguments usefully. If a placement is only worth buying when the link is unqualified, you were never buying advertising. You were buying a ranking signal, and now you know what to call the line item.
The same logic covers gifted product, which the policy names directly. Send someone a product in exchange for a write-up containing a link and the link needs qualifying, whatever the arrangement is called internally.
Corrections
Reading the source instead of the summaries is worth doing partly for what turns out not to be in it. Four things get quoted as policy constantly and are not policy at all.
No toxicity score, no DR threshold, no safe percentage, no absolution through nofollow. All four are inventions.
There is no toxicity score, whatever the vendor glossaries imply. No Google document uses the term toxic backlink, and no ranking system publishes one. It is a vendor metric, computed differently by every tool that sells it, and it is a reasonable prompt to go and look at something. It is not a verdict, and disavowing a list because a tool colored it red is how sites remove links that were doing them good.
There is no domain rating threshold. DR and DA are third party estimates, not Google metrics, and no bullet in the policy references authority scores in any form. A link from a strong site can be a violation and a link from a weak one can be perfectly legitimate.
There is no safe percentage of anything. Not for reciprocal links, not for exact match anchors, not for paid placements. Google has never published a ratio, and every specific number you have read was invented by the person who wrote it.
And there is no rule that nofollow launders intent. Adding attributes to a program of manufactured links makes the individual links compliant and leaves you running a program that no longer does the thing it existed to do. The attribute is a disclosure, not a cleaning agent.
Side by side
Both columns end with a link on a page. Everything that decides which one you are looking at happens before the link exists, which is why auditing links alone rarely settles anything. The three way version of the same problem is what ABC link exchanges actually cost.
| Factor | Editorial link | Link scheme |
|---|---|---|
| Why the link exists | The destination was useful for the point being made | The destination needed a ranking signal |
| Who decided | The publisher, independently | Whoever paid, traded or built the site |
| Relevance | The subjects genuinely overlap | The domains share a metric, not a topic |
| Anchor text | Whatever fits the sentence | Specified in advance, often commercial |
| Placement | Inside the paragraph that needed it | A list, a footer, a partners page, an inserted sentence |
| Scale | Whatever the work earned | Whatever the target required |
| Value to a reader | They click it and get something | Nobody was ever going to click it |
| What happens if you stop | The links keep arriving, because the asset keeps working | Acquisition stops the same month the budget does |
| Explaining it publicly | Straightforward, and often already public | The reason the third hop was suggested |
The last row is the most useful test on this page. If describing how a link was acquired would embarrass you in front of the client, the reader or the publisher, you already have your answer.
Direct answer
No, and it is worth being clear about this because the policy language makes people nervous enough to conclude that any deliberate effort is suspect. Deliberate is not the problem. Nobody at Google has ever suggested that publishers should sit and wait to be discovered.
The test that runs through every bullet is editorial independence. Did the person who published the link decide, freely, that it belonged there? Outreach preserves that decision, because the answer can be no. Payment for unqualified credit removes it. Building the linking site yourself removes it entirely, since there was never a second party to decide anything.
That is why pitching a journalist your dataset is fine and buying a placement on the same publication is not, even when the resulting link looks identical in an export. The link is the same. The decision behind it is not, and the policy is written about the decision.
So the legitimate methods are the ones where you can lose. If nobody can say no to you, you are not doing outreach.
The gray area
Five practices people argue about constantly. None is a violation by default, and each has a specific point where it becomes one. The point is usually scale or payment, which is the same answer the policy gives.
These are ordinary, and treating them as forbidden costs you links you should have.
Reciprocal links
Two sites linking to each other. The policy names excessive exchanges, never reciprocity on its own. Partners, suppliers, associations and integration providers reference each other constantly and always have.
Guest contributions
An article you actually wrote, for a publication that would have run it on merit, with an anchor that fits the sentence. That is a byline, not a placement.
Sponsored placements
Named as acceptable in the policy when qualified. A sponsorship carrying rel="sponsored" is compliant, visible, and still worth buying if the audience is real.
Sitewide links
An agency credit in a client footer, a parent company link, a partner badge. One link on many pages is one editorial decision, and the policy targets wide distribution across many sites.
Industry directories
Listings with a standard, an audience and a reason to exist. Chambers of commerce, associations, comparison sites people actually use.
The same practice, past the point where the policy language starts describing it.
Trading to a target
When the count is the point, relevance goes first and the pattern follows. Adding a third site to hide the pair does not change the arrangement, it changes who can see it.
Placements bought by volume
Mass publication on sites that accept anything, with anchors you supplied. The policy names optimized anchor text in guest posts specifically.
Payment without the attribute
The moment ranking credit travels with money, this stops being advertising. Gifted product counts, and the policy says so in as many words.
Footer links across a network
One agreement producing thousands of links across dozens of unrelated sites. Named in the policy as widely distributed footer and template links.
Directories that refuse nobody
Automatic approval, a fee, no editorial standard, no readers. The policy calls these low quality directory or bookmark links.
The closest call in the table is reciprocity, and it gets its own treatment in are reciprocal links bad for SEO.
The other direction
Go back to the definition. Links to or from a site. Two words most summaries drop, and they change what an audit has to cover, because there is a separate manual action called unnatural links from your site and it describes exactly what it sounds like.
There is a manual action for what you link to, not only for what links to you. Most audits check one direction.
Two of the policy bullets are outbound by nature. Partner pages built exclusively for cross-linking are your pages. Text advertisements that do not block ranking credit are on your site, sold by you. Add advertorials you published and sponsored posts from three years ago, and the outbound half of most link profiles has more unqualified commercial links in it than the owners expect.
This is also the half that a change of staff destroys. Somebody sold a sponsorship in 2022, the post went live with a plain anchor, and the person who agreed it left. Nothing about that link looks unusual to anybody currently at the company, because nobody currently at the company knows money changed hands.
The fix is dull and quick. Crawl your own external links, sort by destination, and find the ones that were commercial arrangements. Add rel="sponsored" where money or product was involved. It is an afternoon, it needs doing once, and then it needs a rule so the next sponsorship ships with the attribute already on it.
Detection
Some of it is trivially detectable and always has been, which is the case for qualifying a domain before you pitch it. Finding every pair of sites that link to each other is a routine query against a link graph. So is clustering sites by hosting, by template, by registration pattern, by publishing rhythm, by the anchors they hand out.
Schemes are not caught one link at a time. They are caught because a hundred links share something a hundred independent decisions would not.
Google has also said its SpamBrain systems are used to neutralize unnatural links rather than only to penalize the sites involved, which describes a mechanism running continuously rather than an investigation somebody opens. That is the part worth planning around: the common outcome is not an accusation, it is that the links quietly stop counting.
Nobody outside Google knows where the internal thresholds sit, and this page is pattern reasoning against published policy rather than inside knowledge. What can be said with confidence is that schemes are almost never caught one link at a time. They are caught because a group of links shares something no group of independent decisions would.
Which is the argument against every hiding technique ever sold. The effort that goes into disguising an arrangement is effort that produced no reader, no traffic and no asset, and the disguise has to keep working indefinitely against systems that get cheaper to run every year.
Footprints
A footprint is a repeated pattern that makes an arrangement legible from the outside, and the link building glossary has short definitions for the ones named here. No single one proves anything, and that is exactly why they get ignored one at a time until four of them stack up on the same set of domains.
Consequences
Penalty is the wrong mental model for most cases, and the everyday version is a link that quietly stops passing anything, and it makes people worry about the rarer outcome while the common one is already happening to them.
Devaluation is quiet and common. A manual action is loud and rare. Most teams worrying about the second are experiencing the first.
Devaluation is the common one. Google's spam systems decide certain links do not count, and that is the end of it. No message, no entry in Search Console, nothing to appeal. The links stay live on the pages, the report still shows them, and the rankings you paid for never arrive. Teams in this position usually conclude their link building is underperforming and buy more of it.
A manual action is the rare one. A human reviewer judges a site to be in breach, the action appears in Search Console with a description, and it stays until you clean up and file a reconsideration request that a person reads. Google's own wording is that this can affect some or all of a site.
The practical difference is what you can do about it. A manual action gives you a defined problem and a defined process. Devaluation gives you nothing to work with except your own honesty about how the links were acquired, which is a harder starting point and the more likely one.
Enforcement
Two versions exist. Unnatural links to your site covers what points at you, unnatural links from your site covers what you point at. These steps follow Google's own recommended actions on its Search Console help page rather than a general cleanup routine.
Google's first instruction, and it is there because most cleanups start from a tool's toxicity score instead of from the policy. The policy is short. Read it against your own profile and you will find the cleanup list is much smaller and much more specific than any tool suggested.
Either arranged by hostname through top linking sites, or in date order through the external links export. Date order is the more useful of the two here, because manipulated links tend to arrive in clusters and the clusters are visible immediately.
Google's guidance for large lists is to start with the sites that link to you most and with links created in the last few months. Judge each one against a policy bullet, not against a vibe. If you cannot name the bullet, it probably is not the problem.
Contact the site owner and ask them to remove the link or stop it passing PageRank with rel="nofollow" or a more specific attribute. Google names both options, and the attribute is usually the easier yes because it costs the publisher nothing.
That is the order Google gives, and it matters. The help page says outright that the disavow tool is often used incorrectly, so the file should be the residue of a real removal effort rather than a substitute for one. Keep the evidence of what you tried.
A person reads it, so it needs three things: what you found, what you changed, and what stops it happening again. If the answer to the third is nothing, the request is a formality that will not clear. Stopping the contract with whoever built the links is part of the answer.
Negative SEO
Far less than the anxiety around it suggests. The systems are built to ignore links they read as unnatural, and links you had nothing to do with are the easiest case there is. If pointing spam at a competitor worked reliably, it would be a cheap industry and it is not.
So the sudden arrival of a few thousand ugly links from sites you have never heard of is usually a non-event, and reacting to it with a large disavow file is a much bigger risk than the links were. That file is a list of links you are asking Google to stop counting, and people building one in a panic routinely include domains that were helping them.
The reasonable response is proportionate. Check the manual actions report. If it is empty, note what arrived, keep watching, and carry on. If a manual action appears, or you find links that somebody working for you actually built, that is a different situation with a documented process attached.
One practical note: you cannot tell whether a placement disappeared, went nofollow or was never live without checking it, and checking hundreds by hand stops happening quickly. That is the job scheduled backlink monitoring does, and it is worth having in place before you need it rather than during the week you do.
Workflow
A working pass, not a full site audit. The aim is not a spreadsheet of everything, it is a short list of links that violate a specific bullet, plus a decision about how they got there.
Search Console's links report is free and it is Google's own view of what points at you. Ahrefs, Semrush or Majestic will give you more history. Deduplicate to referring domains straight away, because a footer link on a 6,000 page site is one decision and it will otherwise look like 6,000.
This is the split that decides everything downstream, and almost nobody starts here. Links you or an agency acquired are your responsibility under the policy. Links that showed up on their own, including the ugly ones, mostly are not. If nobody at the company can say where a batch came from, treat it as built until you find out otherwise.
One exact match commercial anchor is nothing. The same phrase across nineteen unrelated domains is a pattern, and it is the single clearest evidence of how a profile was assembled. Sort anchors by frequency and look at the top ten. You will know within a minute whether you are looking at editorial links.
Mark whether each referring domain is topically related to what you do. It is manual and it is the part that tells you the most. A profile of three hundred domains with forty relevant ones is a completely different profile from the one the count implies.
The domain score tells you almost nothing about the placement. Open the actual page your link sits on. Is it a paragraph somebody wrote, or a list of outbound links under a partners heading? Does the page rank for anything? How many other sites does it link to? Three of those checks take ten seconds each.
Shared hosting, matching templates, identical placement styles, links going live in the same week, the same handful of authors. One coincidence is a coincidence. Four of them across the same set of domains is a network, whether or not anybody involved calls it one.
Crawl your own site for external links and read what you are linking to and why. There is a separate manual action for unnatural links from your site, and old sponsored posts missing a rel attribute are the most common way a clean site fails this half.
Open the manual actions report. If it is empty, you are dealing with devaluation at worst, and the correct response is almost never a mass cleanup. Fix the links you built that violate the policy, add attributes where money changed hands, and leave the random spam alone.
Steps four through six are a screening problem: relevance, placement, outbound counts and whether the site shows any sign of a reader, applied to a list of domains rather than one at a time. That is the part LinkSwapy handles, along with keeping a record of what each partner agreed to, which is the only way to answer how many exchanges you are actually running. See how domain qualification works.
Reading the email
Most link schemes introduce themselves before you agree to anything. The two columns below are the same request, written by people with completely different reasons for sending it.
It refers to something specific you published
Not great content or loved your blog. An actual reference to a piece, with something the sender clearly read. This is the part that cannot be merged from a field, which is why it is the fastest signal available.
The anchor is left to you
Use whatever wording fits your sentence. A publisher is being asked to publish, not to place a keyword, and a sender who means it will say so unprompted.
There is a reason for the link beyond the link
A dataset you would cite, a tool your readers would use, an expert who can answer something your piece left open.
Nothing is conditional
No return link required, no minimum domain rating, no reciprocal obligation attached to the offer.
You could say no without it being awkward
The defining property of an editorial relationship. If declining ends the conversation cleanly, the conversation was legitimate.
A guaranteed number, at a price
One hundred links for fifty dollars, or twenty DR50 placements a month. Nobody can guarantee an independent editorial decision, so a guarantee is a statement about who controls the publishing side.
The exact anchor text is supplied
Usually a commercial phrase, sometimes with a location in it. One line, and it tells you more about intent than the rest of the email.
A spreadsheet of domains arrives with it
Sorted by domain rating, unsorted by subject. The list is the offer, which means relevance was never part of the arrangement.
Nothing in it is about the content
No reference to what either site publishes, because the content is not what is being traded and both parties know it.
The structure exists to hide the structure
A third site is introduced so the trade will not be visible, or you are asked to keep the arrangement off the record. That instinct is telling you what the arrangement is.
Cleanup
The instinct after reading a policy page is to start deleting, and it is the wrong instinct. Work from an inventory of what you actually have first. Stripping legitimate links out of a profile to fix a problem you have not confirmed costs you real signals in exchange for nothing.
Google's guidance is that most sites never need the disavow tool, and its manual action help page says outright that the tool is often used incorrectly. Treat the file as the residue of a real removal effort, not as a substitute for one.
Recovery
Whether you are working a manual action or you have simply worked out that last year's link building was manufactured, the order is the same. The first step is the one people skip, and skipping it makes the rest pointless.
Cancel the retainer, pause the campaign, end the marketplace subscription. A cleanup running alongside continued acquisition from the same source is not a cleanup, and a reconsideration request that cannot answer what stopped it happening again does not clear.
Inherited profiles are the common case, and the person who can explain them has usually left. Reconstruct it from invoices, old campaign reports and the dates in your backlink export. The clusters are usually obvious once you sort by first seen.
Name the bullet or leave the link alone. This single rule keeps a cleanup proportionate, and it is what turns a vague list of suspicious domains into a defensible set of decisions you can put in front of a reviewer.
Contact the site owner and ask them to take the link down, or to add rel="nofollow" or something more specific. Google names both outcomes as acceptable. The attribute is usually the easier yes, because it costs the publisher nothing.
Sponsored posts, partner pages, paid text links, anything commercial that ships without an attribute. There is a separate manual action for this half and it is easier to fix than the inbound one, because you control every page.
Only the links you genuinely could not get removed, and only the ones you or someone acting for you built. Keep the record of what you attempted, because that record is most of what makes the next step credible.
What you found, what you changed, what you could not change and why, and what has been put in place so it does not recur. A reconsideration request is read by a person, and the ones that clear read like an account rather than an apology.
The rankings those links were holding up were never yours, so removing them does not create a gap, it reveals one. Fill it with something that keeps working after you stop paying attention to it. That is the next section.
What to do instead
Every method below shares one property that a trade does not have: the output keeps earning after the campaign ends. A statistics page picks up links for years without another email being sent. A bought placement earns one link and needs the next invoice.
That is the honest case for doing this the slow way, and it is a better argument than the risk one. Most people who buy links are not worried about a manual action. They are worried about the quarter. The reason to stop is that the alternative compounds and the trade does not.
None of it is fast, and anyone telling you otherwise is selling something. The first genuinely linkable asset most teams publish takes longer than the entire link buying process it replaces, and then it outlives it by years.
The operational half matters more than people expect too. Outreach at any scale falls over on record keeping rather than on writing: who was contacted, what was promised, what went live, and what is still live now. That is the part an outreach workflow with a real record exists for, and it is also what stops a team quietly drifting back into trading because trading is easier to track.
The test
Run these on the opportunity in front of you rather than on your profile as a whole, alongside the mechanical checks. Two or three uncomfortable answers is a conversation. Five is a decision.
The verdict
Links created primarily to move rankings rather than to help anybody. Not a list of forbidden tactics, and not a threshold you can stay under. Google's policy is one sentence about purpose with eleven examples underneath it, and the examples are mostly descriptions of payment and scale. Which leaves a gradient rather than a rule.
Build something worth citing, then tell the people who would cite it. Every other method on this page is a way of skipping that step and paying for it later.
Questions
Any link arrangement built primarily to manipulate search rankings rather than to help a reader. Worth knowing that Google itself has mostly retired the phrase: the current spam policies call the section link spam and define it as creating links to or from a site primarily for the purpose of manipulating search rankings. The older term survives in Search Console's manual actions documentation, which still describes participating in link schemes.
Google's current name for the same thing, and the definition in the policy is a single sentence about purpose. Everything under it is examples rather than rules, which is why arguments about whether a specific tactic is allowed usually go nowhere. The tactic is not the test. Why the link exists is the test.
The policy lists eleven practices, including buying or selling links that pass ranking signals, excessive link exchanges and partner pages built for cross-linking, automated link creation, requiring a link in a contract without allowing a qualifying attribute, text links that do not block ranking credit, advertorials that pass credit, low quality directory and bookmark links, keyword-rich links in widgets distributed across sites, widely distributed footer and template links, optimized links in forum comments and signatures, and creating low-value content mainly to manipulate linking signals.
Only past a point the policy does not quantify. The bullet says excessive link exchanges, and it gives partner pages built exclusively for cross-linking as the example. Two relevant companies linking to each other because it made sense for readers is not what that describes. A quarterly quota of swaps sorted by domain rating is.
Not by themselves. Reciprocity describes direction and nothing else, and the policy never names it on its own. What it names is scale, plus pages whose only purpose is holding the trade. If the same two sites would still reference each other with links switched off, there is nothing in the policy that touches it.
The third hop is not itself a violation, and it is also not the shield it is sold as. It removes one signal, the visible A to B pair, and leaves the anchors, the placement style, the timing and the relevance exactly where they were. If the arrangement would be a scheme with two parties, adding a third does not change what it is, it changes how hard it is to query.
Only when they pass ranking credit. Google's policy says directly that buying and selling links is a normal part of the economy of the web for advertising and sponsorship, and that such links are not a violation as long as they are qualified with rel="nofollow" or rel="sponsored". The payment is not the problem. The unqualified ranking credit is.
Not as a format. The policy names something narrower: links with optimized anchor text in articles, guest posts or press releases distributed on other sites, and advertorials where payment buys links that pass credit. A contributed article you actually wrote, for a publication that would have run it anyway, is not what that bullet describes.
In practice yes, though it is worth knowing the policy never uses the term. A private blog network is covered by the bullets it triggers: links built primarily to manipulate rankings, sites bought or run for the purpose, and low-value content created to move linking signals. The absence of the acronym is not permission, it is just a reminder that Google writes policy about behavior rather than about the names SEOs give things.
Finding repeated patterns across a link graph is routine work: shared anchors, shared hosting, shared templates, links going live in step, sites that link out constantly and rank for nothing. Google has also said its SpamBrain systems are used to neutralize unnatural links rather than only to penalize sites. What gets caught is rarely one link. It is a hundred links that share something a hundred independent decisions would not.
Two outcomes exist and only one is a penalty. The common one is devaluation, where the links stop counting, nothing appears in Search Console and rankings simply do not arrive. The rarer one is a manual action for unnatural links, which does appear in Search Console, involves a human reviewer, and needs a cleanup and a reconsideration request before it clears.
A manual action issued when a reviewer finds a pattern of artificial or manipulative links. There are two versions. Unnatural links to your site covers what points at you, and Google's guidance is to export your links from Search Console, get the violating ones removed or qualified, disavow only what you could not remove, and then file a reconsideration request. Unnatural links from your site covers what you link out to, which is the half most audits never look at.
Usually not. Google's own position is that most sites never need the tool, and its manual action guidance explicitly notes the disavow tool is often used incorrectly. Reach for it when you have a manual action, or a documented history of links you or a previous agency built that you cannot get removed. Random spam pointing at your site from sites you have never heard of is not that.
Most of the time they do nothing, because the systems are built to ignore them rather than to punish you for what strangers do. The real risk sits with links you had a hand in: bought placements passing credit, networks, scaled exchanges. That is also the distinction to hold on to when a tool shows you a toxic score, since toxicity is a vendor metric and not something Google publishes or uses.
Build the thing first. Original data, a free tool, a genuinely useful resource, real expertise offered where it belongs, and then tell the people who would want it. The test that survives every policy rewrite is whether the publisher made an independent editorial decision. If they did, no bullet in the policy applies. If they did not, the format you wrapped it in does not matter.
Keep reading
The specific arguments this page summarizes, plus the operational half: screening partners before you agree, and knowing what is still live afterwards.
Link Exchange Guidelines: Domain, Placement, Anchor, and Quality Standards
Most teams have standards, in one person's head, and they hold until volume outgrows that person. Here is the version you can hand to somebody else: eligibility, domain and page criteria, anchors, and the verification step nobody writes down.
Read moreLink Exchange vs Listicle Exchange: SEO Value, Risks, and Which to Use
Listicle exchanges get sold as the natural-looking version. They are also the only exchange format where the page makes a factual claim about merit, and where the two sides are almost never trading assets of equal value.
Read moreAre Link Exchange Platforms Safe? How to Judge One Before You Join
How a platform makes money predicts what it will push you toward, more reliably than anything on its homepage. The five models, the incentive baked into each, and what to ask before you hand over a domain.
Read moreBulk domain quality checker
Screen a list of proposed partner domains in one pass, before anyone replies to the email.
Read moreBacklink monitoring
Scheduled rechecks that flag removals, nofollow flips and noindex changes on your placements.
Read moreLink building glossary
Plain definitions for link spam, manual actions, disavow files, footprints and the rest of the vocabulary here.
Read moreTrack every partnership, what each side committed to, and whether the link is still there today. No marketplace, no automated placement, no domain lists to trade from.