Skip to content
LinkCensus

A dated view of the web's link neighborhood.

LinkCensus uses open Common Crawl web-graph data. Each edition is dated and immutable.

COVERAGE
EDITION 2026-01
Source
Common Crawl host graph, Nov 2025 to Jan 2026
Coverage
52.9M hosts with at least 8 links, 15.6M searchable
Not covered
Small or new domains may be missing.
Cadence
New editions may be added when built and validated, but no cadence is promised.

How a neighbor is found.

ILLUSTRATIVE EXAMPLE

Subject domain

coffee-notes.example

Host embedding

Candidates ranked by graph proximity

  1. brew-guides.example91
  2. roaster-directory.example, compared on the right87
  3. espresso-forum.example84
  4. coffee-gear-review.example79

Ranked by graph proximity, scaled 0 to 100

Linkers of coffee-notes.example, showing 5 of top 200 retained

  • beanwire.example (shared)
  • cuppingroom.example (shared)
  • kettle-weekly.example (shared)
  • roastlog.example
  • moka-daily.example

Linkers of roaster-directory.example, showing 5 of top 200 retained

  • beanwire.example (shared)
  • cuppingroom.example (shared)
  • kettle-weekly.example (shared)
  • grindsize.example
  • tamp-talk.example

19 shared linkers (among top 200 retained)

The direction of a link is ignored in the host embeddings. Shared linkers are retained from the strongest 200 available for each indexed domain, so a count is a floor, not a total. The diagram shows five linkers per column for legibility.

Built on Common Crawl, sold as results, never as data.

The data comes from the Common Crawl web graph and host embeddings, used under the Common Crawl Terms of Use. That license cannot be passed on, which sets the boundary of what LinkCensus offers.

What LinkCensus sells

Access to search results and neighborhood reports: ranked related domains with the shared-linker evidence behind each one, plus exports of those result rows.

What LinkCensus never does

It never redistributes the host embeddings, the web graph, the host table or the search index, in any format. Exports and reports contain result rows only. It never runs a language model over crawled content; nothing in the product is generated by an AI model.

Retained linkers

For each indexed domain the edition keeps its strongest 200 in-linking domains. A shared-linker count is the overlap of two such lists, so it is a floor, never a total. That is why the interface always says "shared linkers (among top 200 retained)".

Report notes

The one-line note on each report row is a deterministic sentence assembled from that row's own data: the shared-linker count, up to two named linkers, whether the TLD matches, the rank band and a hub flag. The same data always produces the same note.

Domain owners can have a domain suppressed from results, exports and reports after verifying control.

Read the opt-out policy

Start with a domain you know.

Free preview: first five neighbors, no sign-up.