Half the linked web has a single referring domain. The most-linked one has 17.7 million.

Half the linked web has one or two backlinks

This study comes down to two numbers. In the March–May 2026 Common Crawl web graph, the median domain that gets linked to at all has two referring domains. The most-linked domain, googleapis.com, has 17,683,579.

That is the whole range. Most sites are linked to by one or two other domains. A few hundred are linked to by millions. Very little sits in between, and almost nothing moves from one end to the other.

We counted every domain-to-domain link in the graph: 4.34 billion links across 118.8 million domains. This post answers three questions. Which domains the web links to most, what those domains actually are, and how far the numbers drop once you leave the top.

How we counted

The data is the Common Crawl web graph, March–May 2026 release (cc-main-2026-mar-apr-may). It is the same open, domain-level link graph Linkwake runs on. Every node is a registrable domain. A link from domain A to domain B means at least one page on A linked to at least one page on B during the crawl. This release has 118,760,321 domains and 4,343,610,896 links.

A domain's referring domains is its in-degree: the number of distinct domains that link to it, counted once each. This is the number every backlink tool leads with. When we say googleapis.com has 17.68 million referring domains, we mean 17.68 million separate domains each carry at least one link to it.

Two caveats. This is one release, and Common Crawl is a large sample of the web rather than a full census, so every count here is a floor. And this is a domain-level graph: it records that A links to B, but not whether a person wrote that link or a script inserted it. Every link counts the same. That last point explains most of what the top of the list looks like.

The 25 most-linked domains

Here is the top of the graph. If you expected the most-linked domains on the web to be news sites, encyclopedias, and universities, the list looks nothing like that.

The number-one domain is googleapis.com, Google's host for fonts, maps, and API clients. 17.68 million domains link to it because their pages load a Google Font or a Maps widget, not because anyone recommended it. At number five is googletagmanager.com with 11.17 million, a tag loader no visitor ever sees. Social networks fill most of the rest of the top ten.

The strangest entry is number seven. gmpg.org has 8,459,372 referring domains. It is a dormant microformats project from the early 2000s. It ranks seventh on the whole web because WordPress themes put a profile link to gmpg.org/xfn in the page head, and about a third of all websites run WordPress. Eight and a half million domains link to a spec nobody maintains, because their theme does it automatically. It outranks Wikipedia, which sits at number 22.

Most of the top 100 is infrastructure

We sorted the 100 most-linked domains by hand into four groups: infrastructure (CDNs, fonts, analytics, tag managers, avatar and asset hosts, ad and consent scripts), platforms and SaaS (site builders, hosting, dev and payment platforms, browsers, portals), social and media networks, and genuine editorial or reference sites. The counts, shown in the bar above, are 35 infrastructure, 27 platforms, 28 social, and 10 editorial.

So 90 of the 100 most-linked domains are things sites load or belong to automatically. Only 10 are the kind of site you would email to ask for a link: Wikipedia, a few governments and standards bodies, a handful of major news sites.

This is the limit of a domain-level count. googletagmanager.com's 11 million referring domains and a major newspaper's referring domains get added up the same way, even though almost none of those 11 million domains chose to link. A script tag did. A referring-domain total mixes links a person placed with links a template inserted, and at the top of the graph the templates win. That is why we put so much work into page-level evidence: what the link is, where it sits, and whether it is followed.

Below the top, the numbers collapse

Under the top few hundred domains, the graph drops off fast. Of the 101,568,637 domains that have any inbound link, the median has 2 referring domains and the 90th percentile has 12. Ninety percent of the linked web is linked to by 12 domains or fewer. You do not reach 509 referring domains until the 99th percentile.

Grouped into tiers, the shape is a cliff. Nearly half of everything with any link has exactly one.

• 1 referring domain: 49,899,751 domains (49.1% of everything with any link)
• 2 to 9: 39,419,052 (38.8%)
• 10 to 99: 9,890,101 (9.7%)
• 100 to 999: 1,632,505 (1.6%)
• 1,000 or more: 727,228 (0.7%). Only 334 domains on the whole web pass 100,000.

The average is 42, and almost nobody is average

The mean number of referring domains per linked domain is 42.8. The median is 2. When the mean sits 21 times above the median, the average stops describing anyone. Real domains have 2 referring domains or they have 2 million. Very few have 42.

The ends carry the story. 49.9 million domains have exactly one referring domain. Another 17.2 million domains, 14.5% of everything in the graph, have zero inbound links at all: they exist and link out, but nothing links back. The most-linked domain has 17.68 million referrers, 8.8 million times the median.

A single authority score, presented as one scale, hides this. There are three regions in practice: a tiny top almost nobody reaches, a wide middle where sites actually compete, and a large floor where most of the web sits on one or two links.

The top is concentrated, and so is the floor

The 10 most-linked domains together hold 116.3 million of the graph's 4.34 billion links: 2.68% of every domain-to-domain link on the web, held by ten domains. The top 100 hold 4.91%. The top 1,000 hold 6.68%. googleapis.com alone accounts for 0.41%.

Read the same numbers the other way. The top 1,000 domains hold under 7% of all links. The other 93% is spread across a tail so wide that its size, not any single giant, carries most of the graph. 727,228 domains have a thousand or more referring domains each. That band, not the top ten, holds most of the real link volume.

So the concentration runs both ways. It is heavy at the very top and heavy at the floor, and fairly ordinary in the middle. The middle is the only part outreach and content can move, and it is the part a single authority number averages away.

The top barely changes between releases

We compared the top 100 domains from the two most recent releases, October–December 2025 and March–May 2026. 98 of the 100 kept their place. Turnover at the very top of the web, over two releases, is 2%.

The two new entries were trustindex.io, a reviews widget, and website-files.com, Webflow's asset host. The two that dropped off were jimdo.com and youtube-nocookie.com. Both changes are infrastructure and platforms trading places as their embed footprints grow or shrink. Nobody entered the top 100 by earning links. They entered by being wired into more sites.

The long tail behaves nothing like this. In an earlier study we found that 44% of the domains linking to github.com changed between a single pair of releases, while its total barely moved. The top of the graph holds still because its links are structural. Boilerplate does not churn.

What this means if you build links

Three things follow, and they cut against how most backlink dashboards get read.

First, the total is a weak comparison. If a competitor shows 40,000 referring domains and you show 4,000, a large share of both counts is infrastructure that neither of you earned: the WordPress link, the CDN, the analytics beacon. The number that separates you is the editorial slice, which is a fraction of the headline.

Second, the reachable target is the middle band. You will not join the 334 domains above 100,000, and you do not need to. The 90th percentile of the linked web is 12 referring domains, and the 99th is about 509. Going from a dozen real editorial referrers to a few hundred puts you ahead of 99% of everything Common Crawl sees.

Third, track which domains you gain and lose, not the total. The top is fixed and the tail churns, so the signal is in the specific domains that changed, and the total hides them. That is what Linkwake reports each release: new, lost, and persistent referring domains, plus page-level evidence for the ones worth checking. Most of a referring-domain count is inherited, and most of it is infrastructure. What matters is which real domains link to you, and whether a person chose to.