Link exchange guidelines
Not automatically, no. Two relevant sites linking to each other happens constantly on the open web, and Google has never said that reciprocity by itself is a violation. What its spam policies name is narrower: excessive link exchanges, the repeated link to me and I will link to you arrangement, and partner pages that exist for the sake of cross-linking.
So the question is not whether a link points back. It is whether a pattern of links exists mainly to move rankings. That is the line, and the rest of this page is about working out which side of it you are on, with a scoring framework at the end for the calls that are genuinely close.
The short answer
Quick answer
Same question, eight different situations. This is the whole argument in one table, and everything below it is the reasoning.
Two relevant sites reference each other because the content warranted it
Ordinary citation. The reciprocity is a coincidence of two separate editorial decisions.
A genuine business partnership both sides describe publicly
Integration partners, suppliers, association members, sponsors.
An occasional editorial link back to someone who cited you
A repeated you link to me, I link to you arrangement
This is the exact behavior the policy quotes. Volume decides how much it matters.
Hundreds of reciprocal placements across your backlink profile
A page built mainly to hold cross links
Partner walls, link directories, friends of the site lists.
Automated or marketplace driven link swapping
Automated link creation is named separately in the same policy.
Paid links that pass ranking signals
A different bullet in the same policy, and a straightforward violation without a rel attribute.
Risk here means how likely a pattern is to be devalued or actioned, read against Google's published policies. 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
It is long because the honest answer has conditions attached. Jump to whichever question you arrived with.
Definition
A link exchange is any arrangement where one site links to another and gets a link back as part of the deal. It can be agreed in a two line email, brokered by a platform, built into a partner program, or routed through a third site to hide the loop. The mechanism changes. The trade does not.
01
Reciprocal link
Two sites that link to each other, recorded in something like the free link exchange tracker. This describes direction and nothing else. It says nothing about why either link exists, which is why the term causes so much confusion.
02
Direct exchange
A links to B, B links to A, agreed in advance. The most visible shape, the one people worry about, and the one Google's policy quotes almost word for word.
03
Three way or ABC exchange
A links to B, B links to C, C links back to A. Same trade with an extra hop, sold on the promise that the loop is invisible. A loop is still a loop.
04
Cross-linking partner page
A page whose job is to hold outbound links to partners. Fine when the partners are real and a reader would want the list. Named as link spam when the page exists to house swaps.
The policy text
Quoted in full
Excessive link exchanges (Link to me and I'll link to you) or partner pages exclusively for the sake of cross-linking.
That single bullet is the entire basis for every argument about this topic, whichever of swap or exchange the email used, and it is worth reading slowly, because two words are carrying the weight. Excessive, not any. Exclusively, not partly.
Google is describing a volume of trading that has stopped looking like citation, and a page whose only reason to exist is the swap. Neither describes a company that links to its integration partner from the integration docs.
The same section names other shapes of the same problem: buying or selling links that pass ranking signals, using automated programs to create links, requiring a link in a contract without allowing a qualifying attribute, and low quality directory links. The policy sums all of them up the same way, as links intended to manipulate rankings.
That framing is more useful than a rule, and it is deliberately durable. Whatever the tactic is called this year, if it exists primarily to move rankings through links, it is in scope.
The qualifier
People read that policy bullet, see the words link exchange, and stop. The qualifier is the whole point. Google did not write no link exchanges. It wrote excessive link exchanges, and then gave an example of the arrangement it meant.
Excessive, not any. Exclusively, not partly. The qualifiers are the policy.
This is consistent with how the rest of the spam policies read. They describe behavior at scale and intent behind it, rather than banning shapes of link. Buying a link is named because money changed hands for a ranking signal. Automated link building is named because a machine placed the link rather than an editor. Exchanges are named because past a certain point, trading stops being citation and becomes procurement.
Which leaves you with a judgment call rather than a checklist, and that is genuinely uncomfortable if you are the person who has to sign off on a campaign. The factors below are what we would look at, in the order we would look at them.
Three scenarios
Not automatically, and treating all three of these the same way is how teams end up removing links they should have kept. The direction of the links is identical in every case. Everything else is different.
Two sites end up linking to each other through separate, unconnected decisions. Neither link was negotiated. Neither party knew the other link would appear. If you asked either editor why the link is there, they would describe a reader, not a trade.
An analytics company publishes a dataset. An agency writes about the dataset and cites the source. Months later the analytics company references the agency's follow up analysis, because it is the best thing written about their own data.
Two companies collaborate on something real, then each documents it on their own site. The links appear because a reader following the story needs to get from one half to the other. Reciprocity is a byproduct of the project, not the reason for it.
A joint survey, an integration, a co-hosted event, a shared guide, an interview. The work exists whether or not the links pass anything, which is the part that makes it defensible.
One dofollow link for one dofollow link, repeated across a list of domains chosen by their metrics. This is not citation with a coincidence attached. It is a trade, and at volume it produces a pattern that looks like exactly what it is.
The email that opens with a quick collaboration idea, includes a spreadsheet, specifies the anchor text, and never mentions a single thing about the content on either site.
This page is about the policy: where trading at scale crosses a line. If your question is narrower, about the reciprocal link itself, what it passes, and how to find the ones you already have, we split that out. are reciprocal links bad for SEO.
Consequences
Penalty is the wrong mental model for most cases, and it is why people worry about the wrong outcome. Two mechanisms exist, they behave completely differently, and the common one is silent.
Devaluation is quiet and common. A manual action is loud and rare. Most teams worrying about the second are actually experiencing the first.
The first is devaluation. Google's automated spam systems, SpamBrain among them, are used to neutralize the credit that unnatural links pass. Google described this directly when it announced the December 2022 link spam update: links identified as unnatural stop counting, and sites that were relying on them can see rankings fall as a result. There is no message, no entry in Search Console, nothing to appeal. The links keep existing and stop doing anything.
The second is a manual action, specifically the unnatural links to your site action. That one does appear in Search Console with a description of the problem, it involves a human reviewer, and clearing it means cleaning up and then filing a reconsideration request. It is a much heavier outcome and a much rarer one. Exchanges are one bullet in a longer policy, and if you want the rest of it, what Google's link spam policy actually names walks through all eleven practices.
Which of these you are exposed to is mostly a question of scale and obviousness. A dozen reciprocal links across a real business with real partners does not read like a scheme to an automated system, because it is not one. Four hundred reciprocal placements acquired in a quarter, all with commercial anchor text, from sites with nothing in common, reads like precisely one thing.
Here is the part most pages on this topic will not tell you. Nobody outside Google knows where the internal thresholds sit, and we do not either. Everything on this page is pattern reasoning against published policy plus the judgment that comes from watching link profiles, and dressing that up as certainty would make it less useful, not more.
The line
Rarely one factor on its own, and the same stacking logic runs through every practice the policy names. It is the stacking that does it, and by the time three or four of these are true at once, the pattern is no longer ambiguous to a person, let alone to a system built to find it.
Would this link still be here if Google counted nothing?
01
One reciprocal link with a supplier is a relationship. Two hundred is a program, and a program leaves a footprint. Volume also changes what the individual links mean: the same swap that reads as ordinary at number three reads as procurement at number ninety.
The tell: Reciprocal placements arriving in batches, from unconnected sites, inside the same week.
02
This is the factor everything else is a proxy for. Links exist to send a reader somewhere or they exist to move a signal, and the people involved always know which. The awkwardness people feel when asked to explain a specific link is usually the answer arriving before they say it.
The tell: You can only justify the link by talking about metrics, never about a reader.
03
SEO software linking to an SEO consultant makes sense to a reader. SEO agency to online casino to pet store to roofing company does not, and no amount of careful anchor text repairs that. Relevance is also the first thing quietly abandoned once somebody sets a monthly link target.
The tell: Your partner list has no subject in common, only a metric in common.
04
A link inside a sentence that needed it behaves nothing like a link in a footer rendered on 4,000 pages. Partner walls, resource lists and friends of the site pages announce the trade before anyone reads the anchor, which is exactly why the policy names them.
The tell: The page holding your link has no purpose you could describe without using the word links.
05
Repeated exact match commercial anchors are the loudest unnatural signal a traded profile produces. Real citations are messy: brand names, bare URLs, an article title, the word here, somebody's surname. Manufactured ones are tidy, and tidy is the problem.
The tell: Every reciprocal link you own says some version of the same commercial phrase.
06
Google's policy names automated programs and services used to create links to your site as its own category. A tool that places links for you, or a marketplace that fulfills a swap with no human deciding the link belongs on that page, sits squarely inside that description.
The tell: No person on either side chose the sentence the link went into.
Side by side
Both produce two links pointing in opposite directions. Everything a spam system can measure sits in the columns below, which is why direction turns out to be the least informative thing about a reciprocal link.
| Factor | Genuine relationship | Manufactured exchange |
|---|---|---|
| Purpose | Helps a reader get somewhere useful | Moves a ranking signal |
| Relevance | Same subject, same audience | Whatever had capacity that month |
| Scale | Occasional, tied to real projects | Systematic, tracked against a quota |
| Anchor text | Brand, URL, or whatever the sentence needed | Optimized, repeated, commercial |
| Editorial control | Either side can decline without consequence | Reciprocity is the condition of the deal |
| Placement | Inside the content, where it was earned | Partner page, footer, resource list |
| Partner selection | Chosen for fit and audience overlap | Chosen by domain rating |
| Pattern over time | Irregular, mirrors real activity | Batched, evenly spaced, paired |
Read down the right column and notice how little of it is about reciprocity. Teams running this at scale usually reach it through an agency workflow rather than one deal at a time. Almost every signal that distinguishes a scheme would still be there if the links pointed in only one direction.
Perfectly ordinary
Before anyone starts auditing, it is worth remembering how much of the web links back to itself for reasons that have nothing to do with search. Every pattern below is a mutual link, and none of them is what the policy is aimed at.
The threshold question
There is no published number. Not ten percent, not twenty, not any figure, and every article you have read that quoted one either guessed or repeated somebody else's guess with more confidence attached. We are not going to invent one here to seem more helpful.
Real link profiles are lumpy. Quotas are smooth. Smoothness is the thing that looks wrong.
What you can do instead is look at the shape of your link acquisition and ask whether it resembles the way citations actually accumulate. Real profiles are lumpy. A piece of research picks up thirty links in a fortnight and nothing for a year. A tool page collects one link a month from people who found it. Manufactured profiles are smooth, because a quota is smooth.
The signals below are what we would pull before making a judgment about a specific profile. None of them is decisive alone, which is rather the point.
The question everyone asks
This is the most asked and least answerable question on the topic, and the honest response is that the number is not the variable. What you can measure is whether each partner reciprocated. Fifty reciprocal links across fifty genuine industry relationships is a different object from fifty reciprocal links bought in a Slack group last month, and no percentage distinguishes them.
Nobody has published a safe percentage. Anyone who quotes one is telling you about their content strategy, not about Google's.
The better question is whether your backlink profile reads like citation behavior or like a trading network. That is answerable, by you, today, without any threshold at all.
If you want a working rule, here is ours, offered as judgment rather than data. If you could not explain a specific reciprocal link to the client whose site it points at without mentioning rankings, it should not be in the profile. That test is stricter than anything Google has published and it has the advantage of being something a team can actually apply on a Tuesday afternoon.
And if you are asking because somebody has set you a monthly link target that can only be hit through swaps, the target is the problem, not the ratio.
Detection
Yes, trivially. Finding every pair where A links to B and B links back to A is a routine query against a link graph, and Google has the graph. This has been technically easy for as long as search engines have existed, so any strategy whose safety depends on reciprocity being invisible was never safe.
The more interesting question is what happens after detection, and the answer is that reciprocity on its own is weak evidence. Half the useful web links back to itself. If mutual linking were treated as spam by default, every association directory and integration doc on the internet would be caught in it.
So reciprocity gets combined with everything else: relevance, placement, anchor distribution, timing, the quality of the sites involved, and whether the same participants keep reappearing together. Google says its automated systems, including SpamBrain, handle spam at scale, and the December 2022 link spam update was announced specifically as SpamBrain being extended to neutralize unnatural links.
Practical consequence: stop optimizing for whether the pattern can be seen. It can. Optimize for whether it holds up once it has been.
ABC exchanges
The structure is simple. A links to B, B links to C, C links back to A. No two sites point at each other, so the direct pair that a reciprocity query would find never exists. That is the entire pitch, and it is why three way swaps get sold as the safe version.
If you need the third hop to feel comfortable, that instinct is telling you something true about the arrangement.
It does remove the pair. It does not remove the arrangement. If one coordinator is placing all three links, the sites are connected by the coordination whether or not they are connected by an edge in the graph, and the footprints that actually give schemes away are mostly the other ones: the same placement style on all three, the same anchor conventions, the same handful of neighborhoods, and links going live in step.
There is also a practical cost nobody mentions in the pitch. A three way loop is harder to hold together, because now the partner you gave a link to owes it to a third site rather than to you, and enforcement gets fuzzy fast. Return links that nobody clearly owns are the ones that quietly never appear. The longer treatment, including what happens once a single triangle grows into a pool, is in what ABC link exchanges actually cost.
Our position, stated plainly: routing a trade through a third site specifically to avoid detection is an admission about the trade. If the placements would stand on their own, you would not need the hop. We put the two structures side by side, with the risk factors compared row by row, in ABC versus reciprocal link exchanges.
Bylines
Site A writes for site B, and later site B contributes to site A. Whether that is ordinary industry publishing or a link scheme in a jacket depends on one testable thing: would either piece have been commissioned if no link were attached?
Two practitioners writing for each other's audiences because the audiences genuinely overlap is how trade publishing has worked forever. Editors do this openly. Nobody counts the links.
A quota of posts traded for a quota of links, with the articles written backwards from the anchor text, is a different object. The tell is almost always the content itself. If nobody involved can name a reader who wanted that article, the article was not the point, and everything about the piece will show it: the generic angle, the padded intro, the link that appears in paragraph two for no editorial reason.
Worth saying since we sell software in this space and it would be convenient to be vague about it: we do not run a guest post program on this site, and we decline the pitches. Given what we sell, doing otherwise would be a slightly awkward look.
Platforms and marketplaces
Wrong question in that form. A platform is not safe or unsafe as a category, it is safe or unsafe according to what it makes you do. These are the characteristics we would judge one on.
We build software in this category, so read this next part with that in mind. LinkSwapy tracks partnerships you already have: which partner owes a return link, whether it went live, whether it is still live, and whether it quietly switched to nofollow. It does not sell links, place them for you, or hand you a list of domains to trade with. That is what keeps the tool on the right side of the policy, and it is also a real limitation: it will not go and find you links, and it is the wrong purchase if that is what you wanted.
Reframe
The exchange mindset asks: give me a backlink and I will give you one. It is a transaction, it is finite, and the moment either side stops needing links it ends. Everything about it is fragile, including the links.
Traded links are the first thing a partner removes when they clean house. Nobody on their side is attached to them.
The partnership mindset asks a different question: can we make something together that is worth referencing? The links that come out of that are a side effect, and side effect links are the ones that survive redesigns, editor changes and content audits, because somebody has a reason to keep them.
This sounds like the soft answer, so here is the hard version of it. Traded links are the first thing removed when a partner cleans house, because nobody on their side is attached to them. Links attached to real work are the last. If you have ever watched a batch of swap links evaporate in a single quarter after a site redesign, you already know which kind you were holding.
Work that produces links as a consequence rather than a condition: joint research, industry surveys, expert roundups where the experts are real, integration documentation, collaborative guides, interviews, co-marketing, case studies with named clients, and events.
Practical framework
Seven checks, in this order, before you agree to anything. The order matters: relevance first, because it disqualifies more opportunities than the other six combined and takes thirty seconds.
Not category relevance, subject relevance. A general marketing blog is not the same neighborhood as a link building tool, even though a spreadsheet would file both under marketing. Ask whether a reader on that page would plausibly want what you are linking to. If you have to construct an argument for it, the answer is no.
Open it. Read the paragraph the link is going into. If there is no paragraph, only a list of outbound links with a heading above it, you have learned everything you need to know about what that page is for. Count the existing outbound links while you are there.
Domain rating is a third party estimate of link strength, and it is the single easiest metric in this industry to inflate. Check whether the site ranks for anything, whether the content reads like it was written for a person, whether the archive has a history, and how many unrelated outbound links sit in the footer and sidebar.
Then ask the question that runs through this entire guide. Would this link still make sense if backlinks counted for nothing? If the honest answer is that the request only exists because of ranking value, you are not evaluating a partnership, you are pricing a trade.
A partner who says use whatever wording fits your sentence is behaving like a partner. A partner who supplies exact wording, especially wording with a commercial keyword and a city in it, is behaving like a link buyer. This one line tells you more about intent than the rest of the email.
The same swap can be fine as your third reciprocal link this year and a problem as your ninetieth this quarter. Nobody evaluates this, because each individual link looks reasonable in isolation. Pull the count before you agree, not after somebody asks you about it.
The final test, and the one that settles arguments. Picture a specific person reading that sentence and clicking that link. Do they get something useful on the other side? If the answer requires you to talk about domain authority rather than about the reader, you already know.
Scoring rubric
Score each factor 0, 1 or 2, then add them up. Sixteen is the maximum. The value is not really the number, it is that two people arguing about a domain end up arguing about the same eight questions instead of about instinct.
| Factor | 0Lower risk | 1Watch it | 2Higher risk |
|---|---|---|---|
| Topical relevance | Same subject and audience | Adjacent, needs explaining | Unrelated |
| Editorial value | A reader gains something | Marginal | None |
| Reciprocity | Incidental, nobody asked | Requested | Required as a condition |
| Scale | Occasional | Regular | Programmatic, against a target |
| Anchor text | Brand, URL, or natural phrasing | Partly optimized | Exact match commercial |
| Placement | Inside the content | Resource or partner page | Link wall or sitewide footer |
| Site quality | Real site with a real audience | Mixed signals | Built to hold links |
| Purpose | Reader value | Both, honestly | Rankings only |
Proceed
This looks like a relationship. Place the link where it belongs and let the anchor be whatever the sentence needed.
Review carefully
Something here is being stretched. Find which factor scored worst and fix that specific thing, or decline. Do not average it away.
Walk away
At this point you are not evaluating a partnership. Nothing about the placement will improve once the links are live.
To be explicit, because this is the sort of thing that gets screenshotted without context: this is our internal editorial checklist, not a Google scoring system. No number here maps to anything Google publishes or to any threshold inside its systems. It exists so a team can make a consistent call and write down why.
Attributes
Three attributes qualify an outbound link for Google: rel=nofollow for links you do not want to endorse, rel=sponsored for links that are paid or otherwise compensated, and rel=ugc for user generated content like comments and forum posts. Google treats them as hints for ranking purposes and you can combine them.
If a swap only makes sense when the links pass signals, adding nofollow does not fix it. It just ends it.
The rule that is not negotiable: if money, product, a free subscription, or any other consideration changed hands, the link needs sponsored or nofollow. That is Google's stated expectation for paid and compensated links, and it applies whether or not anyone would ever notice.
What an attribute does not do is launder an arrangement. Adding nofollow to three hundred traded links does not turn the program into editorial linking. It removes the ranking claim from each link, which is exactly the point, and it also removes the reason most people were trading in the first place. If a swap only makes sense when the links pass signals, the attribute was never going to be the fix.
For a genuinely earned reciprocal link between real partners, there is no blanket requirement to nofollow anything, and adding one out of nervousness just discards a legitimate signal. A normal citation is a normal link.
Red flags
Any one of these is a conversation. Three or more in the same offer is a decline, and you can send that reply in one line without feeling rude about it.
Both directions
Abstract principles are easy to agree with and hard to apply. These are the shapes the two ends of the spectrum actually take.
Technology partners
A CRM provider and an integration platform each document the connector they built. Both docs link to the other product, because a customer configuring it needs both sets of instructions.
Industry research
An agency and an analytics provider run a survey together. Each publishes its own write up, and each links to the other half of the work.
Association membership
A trade body lists its members and members display membership. The link is evidence of something verifiable rather than a favor.
Case study and reference
An agency publishes a case study with the client named. The client references the agency on a partners page. Both statements are true and independently checkable.
Joint resource
Two companies produce an industry report together, each hosts a version, and each credits the other. Removing either link would leave a reader stranded halfway through.
A homepage trading ring
A hundred sites trading homepage links with each other. Homepage placement, unrelated industries, everybody linked to everybody. This is the textbook shape and it has been for twenty years.
Enforced one for one
A platform that will not release a backlink until you place one in return. Reciprocity as a mechanical condition is the arrangement Google's policy quotes almost verbatim.
Commercial anchors across unrelated sites
Unrelated domains exchanging exact match commercial anchors. The anchor profile alone gives it away before anyone looks at relevance.
A partner wall
A page created solely to hold reciprocal links, usually titled partners or resources, with no copy on it and no reason for a reader to open it.
A three way network
An ABC structure built specifically so no two participants link to each other. Designed around detection rather than around readers, which tells you what it is.
Strategy
Not as the base of it, no. A backlink profile whose foundation is trading is fragile in two directions at once: it is exposed to a policy change, and it is exposed to every partner who cleans up their outbound links. Both of those things happen.
Exchanges are a consequence of relationships worth having. They are a poor foundation to build on.
Where exchanges do belong is as a consequence. You work with suppliers, integrators, associations, event organizers and researchers. Some of those relationships produce mutual links because a reader benefits. That is not a link building tactic, it is what having a business looks like, and it stays defensible as long as the standards you apply are written down rather than remembered.
The acquisition methods worth building on are the ones that produce links somebody else chose to give you: digital PR, original research, statistics and data pages, tools, expert commentary, resource outreach, genuine guest contributions, broken link opportunities, and unlinked brand mentions. All slower than trading. All of them survive a policy update.
One caveat we should state, since this is a page on a vendor site. We are not neutral here. We sell software for managing link partnerships, and the position above happens to be the one where our product is useful. Judge the reasoning rather than the source.
Comparison
Risk here means policy risk, judged against Google's published spam policies. Quality ceiling means the best outcome the method can realistically produce, not the average one.
| Method | Editorial control | Scale | Policy risk | Quality ceiling |
|---|---|---|---|---|
| Genuine partnerships | High | Medium | Low | High |
| Digital PR | High | Medium | Low | Very high |
| Original research | High | Medium | Low | Very high |
| Guest contributions | Medium | Medium | Medium | High |
| Reciprocal exchanges | Medium | High | Medium to high | Mixed |
| Automated exchanges | Low | Very high | High | Low |
| Paid links that pass signals | Low | High | Named violation | Mixed |
The last row is in the table because people ask about it, not because it is an option. Buying or selling links for ranking purposes is named directly in Google's spam policies. If you take payment for a link, qualify it with rel=sponsored and the question goes away.
Cleanup
Start by not panicking, and specifically by not mass deleting. Stripping legitimate partner links out of a profile to fix a problem you have not confirmed costs you real signals in exchange for nothing.
From whichever backlink tool you use. You need the full list, not the ones you remember building, because inherited profiles always contain surprises and the surprises are the point of the exercise.
Mark every domain you also link out to. This is the set you are actually reviewing. On most sites it is a much smaller number than people expect, and that alone settles a lot of anxiety.
Use the eight factor rubric above, or your own version of it. What matters is that you apply the same questions to every domain rather than judging each one on how you feel about the relationship.
One exact match anchor is nothing. The same commercial phrase across fourteen partners is a pattern, and it is usually the clearest evidence of how those links were acquired, whether or not you were the one who acquired them.
Links a previous agency or a previous team built are the ones worth the closest look, because nobody currently at the company can explain why any of them exist. That is exactly the position you do not want to be in if anyone ever asks.
Keep, remove, or ask for an attribute. Record the reason next to each one. This takes an afternoon and it is the difference between having an answer and having an opinion the next time somebody questions the profile.
Google's own guidance is that most sites never need the disavow tool. Reach for it only for links you or someone acting for you built, that you cannot get removed, and realistically only when you have a manual action or an obvious paid pattern in front of you.
Most of the cleanup above is a data problem: knowing which partners you link to, what you agreed, whether the return link went live, and whether it is still live today. That is the part LinkSwapy handles, and the part a spreadsheet stops handling somewhere around a hundred placements. See how return link tracking works.
The verdict
Reciprocal linking is not the problem. Manipulation is. Google's policies name excessive exchanges and pages built for cross-linking, and both of those descriptions are about purpose and volume rather than about direction. Which leaves a gradient rather than a rule.
Build the relationship first and let the link be a consequence rather than the deal.
Questions
Not by themselves. Google's spam policies name excessive link exchanges and partner pages built exclusively for cross-linking, which is a description of scale and purpose rather than of direction. Two relevant sites that happen to link to each other because the content warranted it are not doing anything the policy describes.
Two different things can happen and only one is a penalty. The common outcome is devaluation, where Google's spam systems stop counting links they read as unnatural. It happens quietly, with no notification and nothing in Search Console. The rarer outcome is a manual action for unnatural links, which does appear in Search Console and needs a reconsideration request once you have cleaned up. Neither is what happens to a site with a few genuine partner links.
Google has never published a threshold, and any specific percentage you have read was invented by whoever wrote it. Judge the pattern instead: topical relevance, where the links sit on the page, whether the anchors repeat, how the partners were chosen, and whether the trade is the only reason each link exists.
They remove the direct pair, which is the easiest thing in the world to query against a link graph. They do not remove the arrangement, and the footprints that usually give schemes away are the other ones: identical placement styles, repeated anchors, and links going live in step. If you feel you need the third hop, that instinct is telling you something about the trade.
They get risky when they are systematic, irrelevant, or built primarily to move rankings. An occasional link back to a company you actually work with is not the thing Google's policy is aimed at. A quota of swaps tracked in a spreadsheet, with partners sorted by domain rating, is.
Yes, and pretending otherwise would be strange. Suppliers, integration partners, industry associations, event sponsors and research collaborators reference each other constantly. Put the link where a reader would want it, let the anchor be the company name, and do not make a return link a condition of the relationship.
Only where something changed hands beyond the content itself. Google asks for rel=sponsored or rel=nofollow on paid or compensated links, and in those cases it is not optional. A genuine editorial reference to a partner does not need an attribute. Adding nofollow to a program of traded links does not convert the program into something else, it just removes the ranking claim that was the point of it.
Finding every pair of sites that link to each other is a routine query against a link graph, so yes, and that has been true for a very long time. What matters is what the reciprocity is combined with. On its own it is weak evidence. Reciprocity plus thin relevance plus repeated commercial anchors plus links going live in the same week is not weak at all.
Keep reading
The operational side of everything above: qualifying partners, holding them to the return link, and knowing the day one disappears.
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 moreLink exchange tracker
Eight deal types modeled properly, with the reciprocity on record.
Read moreSupplier domain qualification
Score a partner site before you agree to the swap.
Read moreLink building glossary
Reciprocal link, three way exchange, cross-linking partner page, defined.
Read moreTrack what each partner agreed to, whether the return link went live, and whether it is still live. No marketplace, no automated placement, no list of domains to trade with.