The backlink market, measured

finding 01

The forgeries cluster where people shop

A domain is counted suspect when its link authority is high but it has effectively no organic search presence: the signature of authority that was bought rather than earned. You would expect that to rise steadily with claimed DR. It does not. It peaks at DR 90–100, where 50.2% of domains fail, and then falls away above DR 70.

That band is not an accident. It is the workhorse range most link budgets are spent in: high enough to be worth buying, low enough not to invite scrutiny. Forgery follows demand.

share of domains flagged suspect · domains measured at right

finding 02

Who checks the inventory matters more than what it costs

Suppliers are not named here. They are grouped by how they operate: a self-serve marketplace lets anyone list a site, while a managed seller has a human decide what gets sold. That one distinction predicts the fake rate better than price does.

Pick a listing at random from a self-serve marketplace and it is 1.1× more likely to sit on a fake domain than one from a curated seller. A domain carried by several suppliers is counted once per supplier here, because that is the choice a buyer actually faces.

Both kinds sell real inventory too, which is the problem: you cannot avoid the fakes by avoiding a supplier, only by measuring the domain.

finding 03

What a link actually costs, by subject

Median price per placement, across every supplier, in USD. Suppliers use unrelated category vocabularies, so these are harmonised subjects rather than any one seller’s labels.

price gap = median difference between the cheapest and most expensive seller of the same placement on the same domain

finding 04

The same domain, two prices

92,012 domains in this catalogue are sold by more than one marketplace at the same placement type. The median gap between the cheapest and most expensive seller is 95%.

Nothing about the link changes. Same domain, same placement, same publisher. The difference is entirely which account you happened to buy through. A buyer with one supplier cannot see this, because from inside one catalogue there is only ever one price.

finding 05

Sites that take gambling money are cheaper, not pricier

Conventional wisdom says restricted verticals carry a premium. In this catalogue they carry a discount, against the median unrestricted placement:

The sites accepting them are not lower quality on paper: their median claimed DR is 31, against 22 for everything else. They are more likely to be fake (29.4% versus 24.1%). A publisher willing to take anything is telling you something, and the market appears to have priced it in.

finding 06

Price and turnaround by placement type

finding 07

Publishers that sell more than they earn

A rating cannot see this; the public web graph can. Of 189,536 sellable publishers with graph data, 11,190 link out to five hundred or more outside domains and to five times as many as link in. The median such site links out to 1,053 domains; the median clean publisher, 10. One is a publication that sometimes sells a placement. The other is a placement business with a publication wrapped round it.

The saturated sites are the pricier ones, in almost every topic. A publisher that has sold a thousand placements has a rating built out of nothing and a price list built on the rating; the market pays a premium for exactly the sites that are worth least. They are flagged suspect here and excluded from plans by default, whatever they charge.

finding 08

What links to the sites that actually rank

For each topic we took the domains ranking in the top thirty for its most competitive keywords (the leaders) and pulled their referring domains. What links to the winners is the closest thing to ground truth a link buyer gets, and it is not what marketplaces are selling: the share of those referrers that any marketplace carries is small, and the ones that do link to many leaders are the ones worth buying. Those publishers are not named here; they are on every quote line as a count, and a plan can be ordered by it.

Every domain in the catalogue carries this as a number, how many of a topic’s leaders it links to, and plans can be ordered by it. A publication links to what ranks. A farm links to whoever pays.

method

How this was measured