What a backlink spam score actually means
48%.
That’s what a backlinks vendor said about my own site’s inbound profile, right after I’d spent two cents a lookup checking three sellers and got 16%, 26% and 44% back.
I opened the disavow tool. Then I did the arithmetic, which took four minutes and should have come first.
It’s a vendor’s model
A spam score estimates how much of what points at you looks automated, low quality or bought. That’s the whole claim.
Google publishes nothing like it. Nothing in Search Console corresponds to it, no support doc references it, and there is no version of this figure that originates from a search engine. A company crawled the web, trained a model on what bad neighbourhoods look like, and printed the output as a percentage because percentages sell.
I pay for that data and I’d pay again. The problem is the percent sign, which makes an opinion feel like a measurement.
What goes into the score
No vendor publishes the formula. You can infer most of it from watching what moves the number.
The heaviest input is the linking domain’s own outbound behaviour. A site pushing eleven thousand outbound links across four hundred pages is a directory whether or not it calls itself one, and everything it points at inherits a share of that. Note that this signal is entirely about them.
Then thin content on the linking side: a hundred words wrapped around a link block, auto-generated tag archives, scraped copies of somebody’s article with the links swapped out.
Then the shape of the link. Exact match commercial anchors repeating across dozens of domains. Sitewide footer placement instead of an in-article link. Batches of domains registered the same month on the same registrar, all pointing at one target.
And whether the linking domain has any audience at all. A site with no visibility that still publishes daily and links out is either brand new or a farm, and the model can’t tell those apart either.
Almost none of that describes your site. It describes your neighbourhood, assembled from decisions other people made.
Fifty domains
Here’s the arithmetic I should have done in the first minute.
My profile at the time: 108 referring domains, 327 total backlinks. Small. 48% of 108 is roughly fifty domains flagged.
Fifty is a number you can audit by hand. So I opened them. Scraper mirrors, a few directory clones running the same template, aggregator sites republishing RSS feeds with the links left intact. None of them bought. None of them evidence of anything but time passing.
Now take those same fifty junk domains and drop them onto the seller who came back at 26% with 7,800 referring domains. Their score moves by a rounding error. Drop their proportional share of junk onto my 108 and I’m in the seventies.
The percentage is largely a function of how big your clean profile already is, which makes it least informative on exactly the young, small sites whose owners lose sleep over it.
Two tools, same site, different answers
Around the same period I crawled that domain for an unrelated reason and counted the links pointing in. Sixteen.
The vendor said 108 referring domains. My crawl said sixteen live followable links.
Both are right by their own definitions. The vendor counts every domain it has ever seen a link from anywhere in its index: dead pages, links removed a year ago, nofollow included. My crawl counted what I could verify that afternoon.
When two honest measurements of one profile differ by that much, the percentage sitting on top of one of them is a statement about an index and very little else.
The site was fine, which is the whole point
The strongest evidence here is boring. That 48% domain was indexed, carried no manual action, had received no message of any kind, and was selling dofollow placements at $250 to buyers who ran their own vetting before they paid.
A screen that flags a healthy property is doing its job. Screens throw false positives by design, because the alternative is missing the real ones. Read the number the way you read a smoke alarm. It tells you to go and look, and most of the time you find toast.
What I nearly did
My first theory was negative SEO. Somebody had aimed a farm at my domain. That story circulates because it’s more interesting than the truth, which is that junk accumulates on anything left online long enough to get indexed.
My second move was the disavow file, and that one could have cost real money.
I had no evidence. I had a percentage and a bad feeling, and I was half an hour away from telling Google to ignore links I had never once opened.
When the disavow file is the right tool
Two situations, and they’re narrower than the tool’s popularity implies:
- you have a manual action and you’re working the specific links it names or implies
- you bought links at scale, you know exactly which ones, and you’re cleaning up after yourself
Outside those, you’re guessing, and the guess runs one way. Links you disavow that were quietly helping don’t come back when you change your mind. The file applies silently. No confirmation, no readout, no way to attribute a later ranking change to it. You can’t measure the outcome, so you can’t learn from it.
Google’s public position for years has been that it ignores most junk links automatically. I don’t take vendor claims on faith and I extend the same scepticism to search engines, but this one is consistent with the incentives. If it were false, negative SEO would be trivial and every ranking on the internet would be a lottery.
Screening a seller before you pay
Now the half that earns its money, because I keep paying for this data.
The first real use is vetting inventory before a transaction. A seller sends you an authority score and nothing else. Two cents and a few seconds gets you their spam score, their real referring domain count and their total backlinks. What that buys is a reason to go and open twenty of their referring domains yourself.
The score doesn’t make the decision. It tells you how hard to look, and the looking is the expensive part.
Price information, not a gate
The threshold habit is where buyers go wrong. Under 30% buy, over 30% walk.
I bought at 44%. Real company, real product, an audience that overlapped mine, and a page where the link made sense to a reader. Their 44% came from being large and public facing for years, the same sediment as mine with far more surface for it to land on.
What the number should change is what you pay. A 40% domain should cost less than a 15% domain, and if a seller won’t move on price once you raise it, you’ve learned how many other buyers they have this month.
A 30% cutoff would also have rejected me at 48%, along with two of the three placements that turned out fine.
The monthly number worth logging
The use almost nobody sets up is watching your own score over time.
48% on its own means very little. 48% in June and 65% in July means an event, and an event is worth twenty minutes of your day. Maybe a popular page got scraped and propagated across a network. Maybe comments were left open on an old post. Maybe a swap partner sold their domain and the new owner monetised it.
A level tells you about your neighbourhood. A delta tells you something happened. Only one of those is actionable.
I log it monthly alongside the outbound dofollow count from a separate audit on the same site, which was 15,321 the last time I looked. Two numbers, one row, thirty seconds.
Those two are related, incidentally. A site with an open outbound surface, comment fields and followable profile links has been advertising itself to precisely the networks that then link back to it. If your inbound spam percentage looks bad and you can’t explain it, check your outbound surface before assuming somebody attacked you.
What it cannot tell you
Here’s the limit that matters most if you buy links. No spam score, no authority metric, no vendor number will ever tell you whether a specific link helped you.
That’s a permanent limit rather than a tooling gap. Rankings move for a dozen reasons at once, most links take months to do anything, and you can’t run the counterfactual, because you can’t unbuy a link and replay the quarter without it.
So anyone telling you a link on a clean domain is worth more than one on a dirty domain is making an inference. It’s a reasonable inference and I act on it every time I buy. It’s still an inference, and the confidence people attach to these numbers is borrowed entirely from the fact that they arrive with two significant figures.
The four minutes that replaced the panic
Before a purchase: pull spam score, referring domains and total backlinks on the seller’s domain. Two cents. Use it to decide how hard to look. Then open the page they’re offering and read what else it already links out to, which no API will do for you.
On my own sites: same number, once a month, written into a row. Chase the jumps, ignore the levels.
What I don’t do any more is reach for the disavow file because a percentage frightened me. That cost me an hour, and the fix was opening fifty domains in a browser and finding out my site had simply been online long enough for other people’s junk to land on it.
The full vetting numbers, and what those purchases actually returned, are here.