Link building statistics: which ones survive a second look
Most published figures arrive without a sample size, a date or a definition. Here are seven kinds of claim, what is wrong with each, and the check that costs a few credits.
LinkBuildingStat research team ·
Myth against reality
A reading guide, not a list of facts about any one study.
The claim
What is usually wrong
The check
Average site has N backlinks
Averages are dragged by giants
Ask for the median
More links means more traffic
Correlation in a table, not a mechanism
Compare your own two dates
Top results have X referring domains
Selected after the fact
Draw your own sample of ten
Links older than a year work better
Survivor bias
Look at lost links too
DR above 50 is good
Scales differ by vendor
Locate it in a niche distribution
Anchor text should be 10 percent exact
A rule of thumb from nowhere
Read your own anchor share
A single link moved us to page one
One story, no control
Treat as anecdote
Why a median beats a headline
Link profiles are skewed. A handful of sites have millions of links and the rest have dozens.
So an average describes almost nobody, and a statement such as the average page has 200 backlinks should make you ask for the middle value and the spread.
Our dr_distribution action, as used in the cost walkthrough too, returns bucket counts, which is the honest form: you see how many domains sit at each score, then place yourself on that shape, and no one has to trust a single number.
Where you stand in a distribution
One credit buys the shape of the whole index. It is not specific to your niche, so use it to calibrate, not to conclude.
It cannot say that a link helped. We count and describe links; the effect on rankings is a separate question that traffic data and Google Search Console answer, badly and slowly.
It also lags. The index refreshes roughly once a month, and Google runs its own updates on its own clock, such as the March 2024 core update and the August 2025 spam update. A dip that matches a date proves nothing alone.
Treat any number as a lead for the next question. When you publish one, give its date and its sample size.
A fairer way to quote us
If you cite our numbers, say 'in our own link index on a given date'. Add the sample if you drew one.
For example: 'Of ten rivals chosen in advance, the median had 165 referring domains.' That sentence is checkable, limited and truthful, which is more than most statistics manage.
Keep a log of figures you quoted. A year on, you will find that a few of them have aged badly, and the log tells you which.
Every figure in a report should be recomputable by someone else with the same index and date, using only what the page states. That is the bar. Few published statistics reach it, and the bar is not high.
Three sentences to be suspicious of
More than 90 percent of pages get no search traffic. You will see this kind of line everywhere, usually without the study, the year or the definition of a page.
Treat it as a prompt to ask who counted, what they counted, and when. If the author cannot say, the figure is decoration.
The same goes for any claim of the form sites with X do Y better, which describes a table sorted after the fact and says nothing about cause.