If you've ever tried to buy backlinks, you know the feeling: marketplaces boasting "150,000+ sites", rows of authority metrics, and no way to tell which listings are real publications and which are shells built to sell you a link. We wanted the actual number. So we measured it.
Method
On August 22, 2026 we mirrored the full public catalogs of two major link marketplaces via their APIs: 159,736 live listings across 141,537 unique domains. Every listing then went through two layers:
- Mechanical screens. English language, dofollow placement, a minimum of real organic search traffic (authority scores without visitors are the classic tell of a manufactured site), domain rating floor, spam score ceiling, a healthy trust-to-citation ratio, and the site itself being properly indexed in Google.
- AI review of every survivor. A vision model visited each remaining site, read it the way a human buyer would (screenshot plus page content), and answered one question: would a real audience visit this site if it never sold a single link? Each site got a 1 to 5 quality grade and junk flags. Total review cost: a few dollars. There is no longer any excuse for marketplaces not doing this.
Results
| Raw listings | 159,736 (two marketplaces, mirrored 2026-08-22) |
|---|---|
| Passed mechanical screens | 4,816 unique domains, about 3% of raw |
| Passed AI review | 3,442 domains, about 2.2% of raw listings |
Of the sites good enough to reach AI review, 28% still failed it. The quality distribution of the survivors: 354 genuinely reputable publications (including sites of major newspapers), 2,174 solid niche sites with real audiences, and 839 real sites that lean heavily on sponsored content. The review is calibrated honestly: selling sponsored posts is what every listed site does and is not itself a fault. Existing only to host links is.
What the rejects look like
The AI flagged, among the reviewed sites: 991 that were broken or hiding behind walls a buyer's crawler can't see past, 475 with thin filler content, 345 link farms, 131 gambling sites wearing respectable metrics, 68 rebuilt corpses of expired domains, and two dozen unmistakable PBN nodes. The recurring pattern is metric washing: authority scores in the 60s and 70s draped over a site no human would read. Metrics can be manufactured; a readership cannot.
Two more findings from the raw data. First, among listings where a marketplace publishes its own indexing statistics, placements take an average of 8 days to enter Google and 13% never index at all, and most listings publish no indexing data whatsoever, so buyers usually fly blind on the one thing that determines whether a purchased link counts. Second, the same site is often listed on both marketplaces at wildly different prices: across thousands of overlapping domains the median gap was 2x. Buyers who use one marketplace routinely pay double.
Why we did this
Because we're building the thing we wished existed: a curated backlink service where every placement comes from that vetted 2%, is described in plain language with a published quality grade, and is verified by the same indexing pipeline this site runs on itself. A backlink that never gets indexed passes nothing; indexing verification is what we already do, measured live on our own pages on the index log.
Before we sell anything, we'll run a pilot on ourselves: buy placements for rank.fast through our own vetted shelf, push them through our pipeline, and publish every timing and outcome here, the same way we published our own indexing race. Terms and guarantees will be published when the pilot's numbers exist, not before.
What this proves, and what it doesn't
It proves the marketplace inventory problem is real and quantifies it: roughly one listing in fifty would survive an honest quality review. It doesn't prove our curation is perfect (grades are a model's judgment, calibrated against human review, and we'll keep publishing how it performs), and it isn't a claim that buying links from the good 2% will rank you: links are one input, and we don't sell rankings. Cohort details: two marketplaces, one snapshot date, English-focused screens. The numbers will drift as inventory changes; we re-screen continuously.