Links, domains and the difference
Two counts describe the same phenomenon and are routinely swapped for each other. A backlink is a single hyperlink pointing at a page. A referring domain is a single site that contains at least one such link, however many it contains in total. How far apart they sit is decided by templates: a link placed in a footer, a sidebar or a navigation menu appears on every page of that site, so one editorial decision can produce fifty thousand backlinks from exactly one referring domain.
Which is why the referring-domain figure is the one worth reading. A profile of two hundred thousand backlinks from forty referring domains describes a handful of site-wide placements. A profile of nine hundred backlinks from six hundred referring domains describes something closer to genuine citation. The first number is easy to inflate and the second is not.
The same profile, two ways
| Count | What it measures | How it distorts |
|---|---|---|
| Backlinks | Every individual link found. | One template placement multiplies by the size of the linking site. Trivially inflated. |
| Referring domains | Distinct sites that link at all. | Treats a passing mention and a considered citation as equal. Still the better number. |
| Referring IPs / subnets | Distinct hosts behind those sites. | Collapses everything behind a large content delivery network into one bucket. |
follow, nofollow, sponsored, ugc
A link can carry a rel attribute qualifying it. The vocabulary is small and the
consequences are frequently overstated.
- No attribute: the default. The link is a plain editorial reference and passes whatever ranking signal the search engine chooses to attribute to it.
nofollow: introduced in 2005 to fight comment spam. Originally a directive to ignore the link; since 2019 treated as a hint, meaning the engine decides.sponsored: the link was paid for or is part of a commercial arrangement. The honest label for advertising.ugc, the link came from user-generated content: a comment, a forum post, a profile field.
What none of these attributes affects is whether the link works. A nofollow link sends real people to the target, appears in referrer logs, and reads to a human exactly like any other citation. The attribute governs one specific machine behaviour and nothing else, and treating a nofollow link as worthless throws away most of what links are for.
The reference-work paradox
Here is the consequence that follows and that almost nobody states plainly. Large reference works,
meaning encyclopaedias, wikis and several major public-sector and academic sites, apply nofollow
to every outbound link as a blanket policy. They do it precisely to remove any incentive to add
links for ranking reasons, which is a defensible editorial position.
The result is that the most trustworthy thing that can happen to a page, a citation in a reference work read by millions and checked by editors, passes no ranking signal whatsoever. Meanwhile a plain link from a dormant forum profile, added by nobody in particular, does. Any honest account of what a link profile means has to hold both of those facts at once: the ranking value and the citation value of a link are different quantities, and they are sometimes inversely correlated.
The practical reading is to score citations twice. Once for what a search engine will do with them, and once for what a human evaluating the page would make of them. A profile whose strongest entries are all nofollow reference citations is in excellent editorial health and unremarkable ranking health, and neither statement contradicts the other.
Anchor text and context
Anchor text, the words inside the link, is the linking author's description of the target, which makes it evidence about how a page is understood elsewhere. A page whose inbound anchors are all its own brand name is understood as a brand. A page whose anchors are all the same commercial phrase is understood, correctly, as having bought them.
Two things matter more than the anchor itself and get less attention. The surrounding sentence tells you whether the link is a citation, an aside or a list entry, and that is usually visible only by opening the page. And the position on the page, body text against footer against a run of links, separates a considered reference from a directory listing. Neither is available from a summary table, which is why link analysis done entirely from exports reaches confident wrong conclusions.
Why link indexes disagree
No vendor observes the whole link graph. Each runs its own crawler, prioritises differently, reaches different depths, and applies its own rules for when a link that has disappeared should finally be dropped from the index. Two tools examining the same target will therefore disagree, and the disagreement is not a defect in either.
Retirement lag is the largest single source of divergence. A link removed from a page today may continue to be reported for weeks or months, because confirming removal requires a re-crawl the vendor has not yet scheduled. Anything time-sensitive read from a link index is therefore reading partly historical data, and the safe habit is to compare one target against itself over time within a single tool rather than to compare tools against each other.
Reading a link profile
Four questions get most of the way. How many referring domains, not backlinks. What proportion are plain editorial links as against template placements. What do the strongest few actually say: read the page, not the row. And how many still exist, which requires checking rather than trusting the export.
Beyond that, treat authority scores as an ordering rather than a measurement: they are computed over these same incomplete graphs and inherit every one of their gaps. And prefer sources that will still answer later: a registration record is published under obligation, whereas a link index is a product, and products get discontinued. Yahoo Site Explorer, the last free public link index, closed in 2011.
