An analysis of 1.3 million backlinks to a twenty-year-old domain where nothing was ever bought — and what it says about link decay, reclamation, and the numbers in your last audit.
Most backlink audits open the same way. Somebody exports a referring domain list, sorts it by an authority metric, flags everything at the bottom as toxic, and recommends a disavow file. It is a tidy deliverable and it answers almost none of the questions a business actually has about its links.
We recently had the opportunity to answer some of those questions properly. My personal site, billhartzer.com, has been publishing since 2006 and has accumulated 1,301,839 backlinks from 22,260 referring domains. Critically, not one of those links was ever purchased — the profile is entirely natural apart from Bill’s own properties.
That combination is rare. It is old enough to have a history, large enough to be meaningful, and clean enough that what it shows is natural link behavior rather than the decay of a bought campaign. We analysed the complete Majestic Historic and Fresh exports, and the results change several assumptions that routinely appear in audit recommendations.
Your links have a half-life, and you should be budgeting for it
The Historic export records a first-indexed date and a lost date for every link, which is enough to build a survival curve rather than an average. Averaging the age of links that have died ignores every link still working and understates lifespan considerably, so we used Kaplan–Meier survival analysis with still-live links censored at their last-seen date.
The median referring domain stops linking after 1,080 days — just under three years. Across the full profile, 49.6% of every link ever recorded is now lost.
For a business, that translates directly into a number that belongs in a marketing budget. If half your referring domains disappear within three years, then maintaining your current position requires acquiring roughly a sixth of your profile annually before you have grown at all. Any link building proposal that models only acquisition, with no replacement assumption, is describing a treadmill as though it were a staircase.
It also reframes what happens when a program is paused. The common expectation is that rankings hold and then slowly drift. What the data suggests is more active than that: the asset base is eroding on a measurable schedule whether or not anyone is adding to it.
Not all links erode at the same rate
Segmenting the survival curve by the Trust Flow of the referring domain produces a relationship we have not seen published elsewhere, and it is monotonic across every band.
| Referring domain | Domains | Median life | Alive at 5 years |
| Trust Flow 0 | 15,421 | 859 days | 22.2% |
| Trust Flow 1–10 | 4,031 | 1,531 days | 40.1% |
| Trust Flow 11–20 | 1,318 | 1,911 days | 52.6% |
| Trust Flow 21–30 | 647 | 2,223 days | 59.5% |
| Trust Flow 31–40 | 397 | 2,375 days | 59.9% |
| Trust Flow 41–60 | 323 | 2,555 days | 60.8% |
| Trust Flow 61+ | 123 | 3,353 days | 69.9% |
A link from a Trust Flow 61+ domain survives roughly four times as long as one from a Trust Flow 0 domain — 9.2 years against 2.4.
This has a straightforward commercial implication. When link acquisition is priced, it is priced on the authority a link passes today. Nothing in the standard pricing model accounts for how long the link will exist. If a strong link both passes more equity and persists four times longer, the true cost-per-year-of-link is far more favourable at the top of the quality range than the headline price per link suggests — and far worse at the bottom.
Put plainly: cheap links are expensive. They are weaker while they last, and they last less than half as long, so the replacement cycle comes round more than twice as often.
Almost no links are lost because a page died
This is the finding with the most immediate practical value, and it contradicts advice that appears in nearly every audit template.
Of 645,202 lost links in the dataset, we examined what the source page was doing at the moment the link disappeared:
| Source page status when the link was lost | Links | Share |
| Fetched successfully — link no longer present | 503,390 | 78.0% |
| Canonicalises elsewhere now | 124,861 | 19.4% |
| 301 permanent redirect | 11,739 | 1.8% |
| 302 / 307 temporary redirect | 4,814 | 0.7% |
| 404 Not Found | 4 | 0.0006% |
| Dead host, DNS failure, timeout | 24 | 0.004% |
97.4% of lost links disappeared while their source page was still perfectly reachable. Four links out of 645,202 were lost to a 404.
Links are not lost when sites die. They are lost when a page is edited, a site is redesigned, a CMS is migrated, or a canonical tag changes. The page continues to exist, continues to be crawled, and continues to rank. The link is simply no longer on it.
What this changes about link reclamation
Reclamation is usually positioned as a marginal tactic — worth a pass once a year, unlikely to yield much. This data suggests it is systematically undervalued, for a specific reason: the prospects are alive.
When a link is lost to a 404, the opportunity is genuinely gone. When it is lost to an edit or a redesign — which is what happens 97.4% of the time — the page is still published, still maintained, and still has an editor who can be contacted. That is a materially warmer prospect than anything a cold outreach list contains, and the pitch is different: you are not asking for a new link, you are pointing out that something was dropped.
In this profile, 717 referring domains at Trust Flow 21 or above have no live link remaining. That is a prospect list most organizations would pay a considerable amount to assemble, and it was sitting inside an export nobody had run.
What this changes about monitoring
A great deal of link monitoring is built around detecting broken links and dead referring pages. On this evidence, that approach would have detected four link losses out of 645,202 across twenty years.
Effective monitoring has to compare referring-domain snapshots over time and flag domains that have dropped out while remaining live. That is a different technical exercise from crawling for 404s, and most monitoring setups are not doing it.
What a genuinely clean profile looks like
Because nothing in this profile was ever bought, it can answer a question that normally cannot be answered: what proportion of a natural link profile is junk?
69.3% of the 22,260 referring domains — 15,421 of them — are Trust Flow 0. Scrapers, aggregators, expired-domain churn, auto-generated directories, comment spam, and sites that republished content wholesale.
None were solicited. All of them accumulated passively over two decades of publishing.
We raise this because a large proportion of the alarm generated by backlink audits rests on the discovery of exactly this kind of domain. A report showing thousands of zero-authority referring domains reads as evidence of a problem — a negative SEO campaign, a legacy of poor vendor choices, something requiring urgent remediation and a disavow file.
It is worth having a calibration point before accepting that framing. Here is a profile in which nothing was ever bought, and 69.3% of the referring domains are Trust Flow 0. That is the baseline. Junk links are what the internet does to anyone who publishes on it.
The same table calibrates vendor claims from the other direction. In twenty years, this domain earned 446 referring domains at Trust Flow 41 or above — approximately 22 per year. A proposal promising fifty links at that quality within a quarter is describing an outcome roughly double what this site achieved annually, compressed into a tenth of the time. That is not an ambitious target. It is a description of something manufactured.
Why headline link counts mislead
One referring domain in this dataset sent 219,159 links — 16.8% of the entire profile. All but a handful carry identical anchor text pointing at a single URL. It is one sitewide link, placed once, counted 219,159 times.
The top 100 referring domains, out of 22,260, account for 84.3% of every link ever recorded. In the Fresh index, 7,395 of 9,772 classified links are duplicates, so a headline figure of 11,815 backlinks describes roughly 2,400 distinct ones.
Any report that leads with a backlink count without stating the referring-domain count, and without separating distinct from duplicate, is reporting a number that can be moved by a factor of five without a single new editorial decision being made.
What we recommend doing with this
- Build a replacement rate into the link budget rather than modelling acquisition alone. Roughly a sixth of the referring-domain base per year is the figure this dataset supports.
- Price links on cost per year of survival, not cost per link. The quality premium at the top of the range buys durability as well as equity.
- Run a reclamation pass before commissioning new acquisition. Lost links overwhelmingly have live source pages and contactable editors, which makes them the cheapest available prospects.
- Change what link monitoring watches. Compare referring-domain snapshots over time; do not rely on crawling for broken pages.
- Establish a junk baseline before acting on an audit that recommends disavowal. A large volume of zero-authority referring domains is normal, and by itself is not evidence of anything.
The full case study — every figure, the complete technical audit, and an interactive survival curve — is published on AdvancedLinkTraining.com, a free nine-module link building course. All figures are computed from Majestic Historic and Fresh exports for billhartzer.com downloaded 2 August 2026. The Historic export returned 1,301,839 rows against the 1,382,639 links reported on the Summary screen. Trust Flow is measured at export time rather than at link acquisition, which biases the lower bands’ survival downward by an unknown amount.
If you would like this analysis run against your own profile, or a second opinion on an audit recommending a disavow file, that is the kind of work Hartzer Consulting does.