Where the numbers come from

The marketplaces

We hold buying accounts with 9 link marketplaces and read their full catalogues by API or, where there is none, by the same pages a customer sees. Every listing is normalised to one schema (placement type, inventory mode, price in USD minor units, turnaround, content rules, the seller’s claimed metrics) and deduplicated by domain and placement type, so a domain carried by four sellers is one thing you can buy, not four.

Prices are observations, not quotes. Each carries the time it was seen; a price older than 48 hours is refused at order time. Every observation is appended to a history that is never overwritten, so the asking price of any listing on any past day is recoverable. That history began on 28 August 2026.

Suppliers are never named in research. They are grouped by how they operate: self-serve, where anyone can list a site and nobody checks it, and managed, where a human curates what is sold.

The backlink index

Every seller quotes an authority score. We measure each domain against a third-party backlink and search index instead: referring domains, a 0–1000 domain rank, spam score, estimated monthly organic traffic and the number of ranking keywords. Traffic and keywords are US English only; a site that ranks mainly in another market reads low here, which is why a low reading alone never makes a domain suspect.

The measured rank and the seller’s claimed DR sit on different scales and correlate at only 0.36 across our sample. We keep both and never substitute one for the other: the disagreement is the signal.

The web graph

A public, domain-level hyperlink graph of the open web: 2.45 billion links between 120 million domains, rebuilt quarterly. From it we take how many outside domains each publisher links to, how many link in, its PageRank over the whole web, and how many of each topic’s top-ranking sites it links to. It is a sample of the web and undated within its release, so it is used to find and to flag, never as the freshest reading. Detail on the web graph page.

Our own orders

What no index has: what happened after a placement was bought. Every delivered link is re-checked at 0, 7, 30, 60, 90, 180 and 365 days: is it there, is it followed, what does the page around it look like. The target page’s rankings are observed at the same checkpoints. Per publisher these become a survival rate; per supplier a scorecard; per order an outcome. All three are published only in aggregate and under floors (three orders for a publisher, three distinct buyers for an outcome), and always as observed after, never caused by.

What is not here

Google's own data

No Search Console or analytics data is used in anything published. When a buyer connects their own, it stays theirs and never joins a shared table.

Clearing prices

We publish asking prices and the spread between sellers. Transaction prices exist only for our own orders and are not yet numerous enough to publish.

Named domains

Except a topic's leaders, which rank publicly for public queries. Nothing about a domain's price, flag or sales is published against its name.