What Are Toxic Backlinks?
Most links a tool calls toxic do nothing at all. The tools flag far more than Google acts on. Acting on the whole list usually does more damage than leaving it alone. If you have just run a report and been alarmed by a score, this page is written for you.
What Toxic Means And Who Decided
There is no Google definition of a toxic backlink. The word is not used in Google's documentation and no toxicity score comes from Google. Every one you have ever seen was produced by a third party tool vendor, using a formula that vendor chose, weighted according to its own judgement about what looks suspicious.
That is worth restating in the form it actually takes on screen. A number appeared next to a link. Perhaps a percentage appeared next to your whole profile. Either way it looked like a measurement. It is an opinion expressed as a number. The vendor had to pick which characteristics count as warning signs and how much each one should weigh. Those choices are commercial and editorial rather than derived from anything Google publishes.
The clearest evidence for this is that the vendors disagree with each other. Run the same profile through different tools and you get different lists, different scores and different verdicts on the same individual link. If toxicity were a property of the link, that would not happen. It happens because each score is a separate guess.
None of which makes the tools useless. They are genuinely helpful for finding things you would never have spotted by hand. The mistake is treating the output as a diagnosis rather than as a list of things to look at. The wider problem of trusting third party metrics is covered in our guide to Domain Authority. The full series sits in the complete guide to backlink building.
Why Tool Scores Overstate The Problem
A scoring model has a built in reason to err towards flagging. It is a commercial reason rather than a technical one. Consider the two ways a tool can be wrong. If it misses something genuinely harmful, a customer eventually complains that the tool let them down. If it flags something harmless, nothing happens. The customer either ignores it or acts unnecessarily. In neither case does the vendor hear about it. One error is visible and the other is invisible, so the sensible product decision is to flag generously.
Then there is what a model cannot see. It cannot tell whether a link was editorially given or arranged. It cannot tell whether the page has readers. It cannot read the sentence around the link and judge whether it makes sense there. It cannot distinguish a genuine small trade directory from something built to host links, because from the outside those can look structurally identical. So it falls back on proxies: how the site looks, how many outbound links it carries, what its own profile resembles.
The last piece is the one that reframes the whole subject. Every website on the internet accumulates junk links passively and continuously, without anybody doing anything. Content gets scraped and republished. Aggregators list you. Automatically generated statistics sites create a page about your domain. Abandoned blogs with spam comments point at you. Sites in languages you do not read copy your pages wholesale.
That is the normal condition of a website rather than a symptom. A site with hundreds of flagged links is unremarkable. A site with none would be genuinely strange. It would probably mean the tool had barely crawled it.
What Google Says It Ignores
Google's published guidance on link spam describes a system that identifies links it does not wish to count then discounts them, rather than one that punishes the site being linked to. Its search representatives have said consistently, in documentation and in public, that the overwhelming majority of sites do not need to do anything about unwanted inbound links, because those links are already being disregarded.
Reading that carefully matters, because two different things get conflated. Discounting a link means it contributes nothing. Penalising a site means it is worse off than if the link had never existed. Google's stated position is that the first happens routinely and the second is reserved for cases where a site has been manipulating its own profile.
For a business that has never bought a link, never used a network and never run an arranged campaign, the practical implication is easy to state. The alarming list in your report is almost certainly a list of links that already count for nothing. They are not costing you rankings. They are not a debt to be cleared. They are the internet's ordinary background noise, correctly ignored by the system that matters and incorrectly flagged by the one that does not.
The situation is genuinely different for a business that did arrange links, which is the subject of the next block. The dividing line is not the appearance of the links. It is whether you caused them.
The Links That Do Cause Harm
Harm attaches to patterns rather than to individual links. It also attaches to intent rather than to appearance. Three categories carry real exposure.
Deliberate schemes. Arrangements set up specifically to produce links at volume, where the participants are all doing the same thing to each other. The individual link may look ordinary. The shape across hundreds of them does not.
Paid networks. Groups of sites under common control that exist to sell placements. What makes these dangerous is not any single site but the fact that they are identifiable as a group, so exposure is shared across everybody who used them.
Anything the site itself arranged. The broadest category and the one that actually decides the question. If your business took steps to make links appear, those links are yours in a way that a scraper's link never is. The same applies to anything a supplier arranged on your behalf.
So the divide worth carrying away is between links you caused and links that happened to you. Links that happened to you are ordinary web debris. A spam blog pointing at your site is not a decision you made and not a signal about your conduct, which is why it is discounted rather than held against you. Links you caused are evidence of intent. Intent is what the guidance addresses.
This is also why a competitor cannot easily damage you by pointing bad links at your site. Doing so produces exactly the pattern of links that happened to you rather than links you caused. Our guide to how Google detects unnatural backlink patterns covers the shapes involved.
When Doing Nothing Is The Right Answer
For most businesses reading this page, the correct action is none. That is not negligence and it is not a shrug. It follows from the three blocks above.
The conditions are easy to check. You have never bought links, used a placement network or run an arranged campaign. Nobody working for you has, including any agency you have used, which is worth actually confirming rather than assuming. There is no manual action notice in your Search Console account. Your visibility has not fallen in a way that coincides with anything.
If all four hold, the flagged links are being discounted already. Removing them changes nothing that was happening. Acting also has a real downside, since every hour spent, every removal request sent and every domain added to a disavow file carries a chance of removing something that was quietly helping you.
What to monitor instead of the score. Non brand visibility, meaning search visibility on terms that do not contain your business name, which is where a genuine problem would show. The manual actions section of Search Console, which is the only place a real penalty is ever announced. Referring domains from sites actually relevant to your field, which is the number worth growing. And whether any drop in visibility lines up with something specific: an algorithm update, a site change, a lost page.
A toxicity percentage belongs in none of that list. Watching it monthly trains a business to manage a vendor's opinion rather than its own performance, which is a slow and expensive habit to acquire.
In What Order To Act
Some businesses did arrange links. Some have received a manual action naming them. In either case the work needs an order in place. Three questions rank it: did you cause it, is it part of a pattern, is removal realistic.
First, links you paid for or arranged directly. Whatever you or a previous supplier commissioned. These are the only links with your intent attached, so they are the only ones where removal changes the picture rather than merely shortening a list.
Second, links that form part of a pattern you created. A group of placements from one arrangement, a set of sites that all appeared in the same fortnight, several links using the same commercial anchor phrase. The pattern is the exposure, so the pattern is the unit of work rather than the individual link.
Third, links from a group you can identify as one you engaged with. If you used a supplier and can recognise the footprint of what they placed, that footprint belongs on the list even where individual sites look harmless.
Everything else: leave it. Scrapers, aggregators, statistics pages, foreign language copies, abandoned blogs, directories you never submitted to. Not yours, not evidence of intent, not worth an hour.
Then apply the realism test. Removal requires somebody at the other end to act. A site that sold placements will often want paying again to remove them, which is a trap covered in the mistakes block. A site that has been abandoned has nobody to ask. Where removal is not realistic, that is what the disavow file exists for. It is the last step rather than the first.
Cleaning Up Without Losing Rankings
Cleanups cause damage regularly. The mechanism is simple enough. Some of the links on a flagged list are passing real value. Remove them and you remove the value with them. Because the loss arrives weeks later and looks like ordinary volatility, most businesses never connect the two.
Establish a baseline before touching anything. Record the referring domains, the total links, the anchor distribution, the pages receiving links, your visibility on non brand terms and the date. This takes an hour. Without it you cannot answer the only question that will matter in two months, which is whether things got better or worse after you started.
Work in stages rather than one pass. Take the highest priority group, act on it, then wait. Weeks rather than days. Then reassess against the baseline before touching the next group. A cleanup executed in a single sweep is uninterpretable: if visibility falls afterwards, every action is equally suspect and none can be reversed with confidence.
Keep a record of what you did and when. Not for anybody else's benefit. For your own, in three months, when something moves and you need to know what preceded it.
The staged approach feels slow to a business that has just been frightened by a report, which is precisely why it is worth insisting on. The alarm is manufactured by a scoring model. The rankings are real. Our guide to auditing your backlinks sets out how to establish that baseline properly.
Where Disavow Fits
The disavow file is a request asking Google to ignore specified links to your site. It sits at the end of the process described above rather than at the beginning. For most readers of this page it does not feature at all.
Its place is narrow. Where a business genuinely arranged links, where removal has been attempted and failed and where those links are the identified problem rather than a suspicion, a disavow file records that position. It is the tool for links you caused and cannot remove.
What it is not is a response to a report. Submitting a file because a tool produced an alarming percentage is the single most common misuse of it. It carries the downside described in block seven with none of the upside, since the links were being discounted anyway.
The detail of what the file does, what it does not do and the narrow set of circumstances that justify one sits in our guide to the disavow tool, which is worth reading in full before creating one.
Common Mistakes
Five mistakes, all of which we have been called in to undo.
Disavowing on a tool score alone. The commonest by a wide margin. A percentage looked bad, so a file went in. The links were already being discounted, so nothing improved. Anything useful in the file is now being ignored on request.
Sending bulk removal requests. Hundreds of identical emails to site owners. Almost all are ignored. A proportion of recipients now know your domain cares about its link profile, which in a small number of cases invites the opposite of cooperation.
Acting before establishing a baseline. Covered above. It converts a reversible experiment into a permanent mystery.
Treating a competitor's links as an attack. Pointing junk at a rival produces links that happened to them rather than links they caused, which is the category that gets discounted. Time spent worrying about this is time not spent on anything that pays.
Paying somebody to remove links they placed. A supplier that sold placements then charges to withdraw them is being paid twice for one piece of work. Where the links are genuinely yours and genuinely a problem, the disavow route does not require their cooperation.
We tell you
when to do nothing.
An audit that ends in a ranked list of actions rather than a toxicity percentage. It frequently ends in the recommendation to leave the profile alone. Where action is warranted it happens in stages against a recorded baseline.
Included every month:
One fixed monthly rate. No setup fee. No price per link.
Twenty four guides.
One subject.
This page covers the alarm. The rest of the series covers what the disavow file actually does, how to audit a profile properly, how patterns are detected and what a third party metric is really worth.
