After this lesson you should be able to
- Distinguish distinct links from duplicates and report the right one
- State the index, date and definitions behind every number you publish
- Show loss alongside gain in every reporting period
- Withstand a hostile reading of your own report
Write for the hostile reader
A link report has more than one audience. There is the person who commissioned the work and wants to know whether it is going well. There is the person who will inherit the account and needs to understand what was done. And in the cases I am most often called into, there is a reader whose job is to find the weakest claim in the document and pull on it.
Write for the third reader and the first two are served automatically. That means every number carries its source, its date and its definition, and every claim you cannot support gets removed rather than softened. A report that says referring domains rose from 412 to 447 between two named dates in a named index is not exciting, but it cannot be taken apart. A report that says we built 200 links is exciting and cannot be defended, because nobody can say what a link means in that sentence.
The habit that makes this easy is to write the definitions section first, before you have any numbers to protect. It is remarkable how much sloppiness that one step removes.
There is a second benefit that has nothing to do with defense. A report built on stable definitions can be compared with itself. Change how you count between March and June and you have destroyed your own trend line, and no amount of commentary will restore it. The discipline that protects you from a hostile reader is the same discipline that lets you learn anything from your own history.
Distinct versus duplicate
Here is the specific mechanism that makes link counts unreliable, with real figures. Taking a Fresh index export of a single profile: of 9,772 classified links, 7,395 were duplicates. Distinct links numbered 2,377. Duplicates were 75.7% of the total.
So the same profile, in the same index, on the same day, can honestly be described as 11,815 backlinks or as roughly 2,400 real ones, depending on nothing but which count you quoted. Apply the index's own noise reduction and the figure lands somewhere else again, near 5,157. Three defensible numbers, a fivefold spread, and not one editorial decision separating them.
What counts as a duplicate varies by index, and it usually means multiple links from the same source page to the same target, or the same link recorded across protocol and host variants. None of that is dishonest on the index's part. It becomes dishonest the moment someone quotes the largest available number without saying which one it is.
The rule: report distinct links and referring domains, name the index, and give the date. If a client or an executive asks why your number is smaller than the one a vendor quoted, that paragraph is the answer, and it is worth having ready before the question arrives.
The numbers a link report needs
A defensible report has a small, stable core. Adding to it is fine; changing it between periods is not, because a report whose definitions move cannot show a trend.
- Referring domains, live, at the period end. The headline.
- Referring domains gained and lost in the period, with both lists attached. Not a net figure with the detail hidden.
- Distinct links, with the index and date named. Raw link totals may appear beside it, labeled as such.
- Quality distribution of what was gained — Trust Flow bands, or your own rubric score, so that 30 junk domains cannot masquerade as a good quarter.
- Relevance judgment on the gained domains, done by hand on a sample if the volume is large.
- What was done — contacts made, replies, placements, and the two ratios. This is what lets a reader tell a good quarter with bad luck from a bad quarter.
One structural note. Attach the raw lists. A report with the underlying domain lists in an appendix invites checking, and inviting checking is exactly what a report that survives scrutiny does. Reports that summarize without evidence are the ones that fall apart when someone finally looks.
Two omissions worth naming. Leave out anything you cannot recompute next period from the same source, because a one-off statistic becomes an obligation the moment it appears in a report. And leave out anything you would have to explain verbally for the reader to interpret correctly. A report is read months later by someone who was not in the meeting, and a number that needs a caveat delivered out loud will eventually be quoted without it.
Show the loss
Most link reports show acquisition only, which makes them descriptions of activity rather than of the profile. Loss is not an unfortunate detail; it is roughly half the story. On the twenty-year profile behind this course, 49.6% of all links ever recorded are lost, and the median referring domain stops linking after 1,080 days.
A report that shows 20 domains gained and stays silent on 18 lost is misleading even if every number in it is accurate. Showing both changes what the reader does with the report: it makes reclamation visible as work worth funding, it explains why a program that looks busy can produce a flat profile, and it sets a realistic expectation for what continued investment buys.
Practically, this needs a dated referring-domain snapshot at every period boundary. Export the list, keep the file with the date in its name, and compute gains and losses by comparing consecutive snapshots. It takes minutes and it is the only method that produces a defensible loss figure, because indexes do not reliably tell you what quietly went missing.
Losing links is normal. Losing high Trust Flow links without noticing is not, and the loss column is what makes the difference visible.
The claims that do not survive
Four things I remove from every report I review.
Causal claims about rankings. Traffic rose after a campaign is a fact. The campaign caused the rise is a claim you cannot support, because you cannot rerun the quarter without it. Show the correlation, name the confounders, let the reader draw the line.
Unsourced quality language. High-quality and authoritative mean nothing without the criteria behind them. If you have a rubric, state it and show the scores. If you do not, say relevant to the subject and leave it there.
Cumulative totals that never decrease. Links built to date is a number that only goes up, which means it is not describing a profile that loses half its links. Report live figures at a date.
Vendor numbers repeated without checking. If a supplier reports placements, verify them: is the link live now, is the page indexed, is it followed, is it on the page they said? In my experience a meaningful share of reported placements fail at least one of those checks, and the report carries your name, not theirs.
Questions
Why is my tool's backlink number so much bigger than my distinct link count?
Because raw counts include duplicates: multiple links from the same page to the same target, and the same link recorded across host and protocol variants. On one real Fresh index export, 7,395 of 9,772 classified links were duplicates, so a profile described as 11,815 backlinks contained roughly 2,400 distinct ones. Both numbers are accurate; only one is informative.
Which link index should I report from?
Whichever you can use consistently, named in the report along with the date. Indexes differ in crawl coverage and in how they classify duplicates, so numbers are not comparable across tools. Switching index between periods destroys your trend line more effectively than any change in performance, so pick one and stay with it.
Should I report links that were lost?
Yes, in every period. Around half of all links ever recorded on a mature profile are eventually lost, and the median referring domain stops linking after about three years. A report showing only gains describes activity rather than the profile, and it hides the reclamation work that is usually the highest-yield thing on the program.
How do I explain a smaller number than the vendor quoted?
Show the definitions. Explain that you report distinct links and referring domains from a named index on a named date, and that raw counts include duplicates and sitewide repetition. The same profile can be described honestly with numbers a fivefold distance apart, so the useful conversation is about which definition is being used, not about which figure is bigger.