Original data · updated 8 September 2026
Link building statistics from 18,560 placements we verified ourselves
Of 18,560 backlink placements we built and later re-checked, 77% were still live. That figure comes from 17 delivered campaigns, and it is worth almost nothing on its own — because the survival rate is decided by what kind of placement it is, not by how carefully the campaign was run. Split by type, the same dataset ranges from every link intact to not one link surviving. The table below is that split, computed from our own delivery records rather than estimated.
Backlink statistics by placement type
Every placement type with at least 20 verified links, worst survival first. A placement counts as live only when a link to the customer’s site is actually found on the page.
| Placement type | Verified | Still live | Survival |
|---|---|---|---|
| Social Bookmark | 265 | 0 | 0% |
| Directory / Citation | 233 | 0 | 0% |
| Doc Sharing | 29 | 5 | 17% |
| Social Profile | 39 | 24 | 62% |
| Web 2.0 | 12,571 | 9,455 | 75% |
| Wiki | 2,338 | 1,789 | 77% |
| Business Listing | 20 | 18 | 90% |
| Backlink | 947 | 880 | 93% |
| Social Bookmarking | 27 | 25 | 93% |
| Article | 2,057 | 2,048 | 99% |
6 further placement types had fewer than 20 verified links and are left out rather than shown at a rate too thin to mean anything.
2 categories produced no surviving links at all
- Social Bookmark — 0 live of 265 verified
- Directory / Citation — 0 live of 233 verified
These are not links that decayed. The pages load normally; they simply carry no link to the customer, and in these categories not one verified placement ever did. They are routinely sold by count inside cheap bulk packages, which is exactly why the count is the wrong thing to buy on.
We found this in our own delivery records, which means we had been passing some of it on before we measured it. Our delivery process now refuses to send a report containing these categories, and refuses any batch that comes back verified and essentially dead, whatever it is labelled as. Publishing the finding is the honest version of fixing it.
Method, and what this data cannot tell you
How it is measured. Every placement in a delivered report is fetched and inspected for a link resolving to the customer’s site. Live means the link was found. Dead means the page loaded and the link was absent. Anything inconclusive is excluded from the percentages rather than counted against the source, so 18,560 of 20,361 placements are represented above. Re-checks run nightly, so these figures move over time.
This is one vendor’s record, not an industry study. It reflects the placement types we sell, the suppliers we use and the sites our customers operate. A different mix would produce different numbers. It is a census of our own work rather than a sample of the market, which makes it complete but not representative.
Live is not the same as valuable. A surviving link on a page nobody reads is still worth little. This measures persistence only — not authority, not traffic, not ranking impact — and a high survival rate should be read as a floor on quality rather than as proof of it.
Some attrition is normal. Self-published properties are removed, blogs close, pages get pruned. A rate in the seventies for that kind of placement is ordinary and priced in. What is not ordinary is a category that never produced a live link, which is why the two findings are reported separately above.
What this means if you are buying links
Ask for the placement type breakdown before the count. A thousand links is a meaningless number until you know how it splits, and the split is what determines how much of it will still exist in six months.
Ask whether anyone re-checks after delivery. Most reports are written at handover and never revisited, which means they describe an intention rather than an outcome. A supplier who re-checks can tell you what survived; one who does not is quoting you the day it shipped.
Treat a claim of total survival as a warning rather than a reassurance. Nothing in this dataset — across 18,560 verified placements — supports the idea that every link in a campaign stays live, and a supplier promising it either is not measuring or is not telling you.
We publish this because it is the same measurement we run before any report goes out. If you want the equivalent read on your own site rather than on ours, the AI Visibility Audit shows where you currently stand in AI answers, and LLM optimization explains how models choose which source to quote.
Questions about backlink survival
What percentage of backlinks stay live?
Across every placement we have built and re-checked — 18,560 verified links from 17 campaigns — 77% were still live when we looked again. That single figure hides the only thing that matters, because survival is decided by placement type rather than by how well a campaign was run. 2 categories — social bookmark and directory / citation — came back at zero across 498 verified placements. Anyone quoting a site-wide survival number without splitting it by type is averaging away the useful part.
How long do backlinks last?
The useful question is not how long a link lasts but whether it was ever really placed. In our data, links that disappear overwhelmingly do so early — the page is deleted, or was published without the link in the first place — rather than decaying gradually over years. An editorial placement inside a real article on a site with its own audience tends to persist as long as the article does. A link on a page created only to hold links tends to vanish with the page, and that is a property of the placement type, not of time.
Are cheap bulk backlink packages worth buying?
Partly, and only if you know which part. Bulk foundation packages are a mix of placement types whose survival rates differ enormously: the strongest components in our data hold up well, while social bookmark and directory / citation measured zero live across 498 verified placements. Those links are real in the sense that the pages exist, and worthless in the sense that no link to the customer survived on them. Buying a bulk pack is therefore not automatically a mistake, but paying for the whole count as though it were uniform is.
Why do most link building statistics come from tool vendors?
Because vendors who sell links have little reason to publish how many of theirs survive, and the companies with the crawl data sell software rather than placements. That leaves a gap: plenty of published research on what correlates with rankings, almost none on whether the links people actually buy are still there afterwards. This page is one vendor's own delivery record, which is a narrower thing than an industry study — see the limits section — but it is measured rather than claimed.
How is this data collected?
Each placement in a delivered report is fetched and checked for a link pointing to the customer's site. A placement counts as live only when that link is actually found, counts as dead only when the page loads and the link is absent, and is left out of the percentages entirely when the check is inconclusive — which is why 18,560 of 20,361 placements are represented. Re-checks run nightly against links already delivered, so these figures move as pages come and go rather than being frozen at handover.