Ghost towns: where the census's dead domains cluster
Top-level domains do not decay at the same rate. In the October 2026 census, the country-code group has the highest dead rate, and three endings explain most of it: .tk, .cf and .gq hold 1,560,141 of the group’s 2,096,902 dead names. New gTLDs, the wave of endings sold since 2013, have a lower rate, and their dead names are concentrated. Ten endings hold more than four fifths of the group’s dead names.
A change in what dead counts
This edition counts dead names on a new basis. It removed every name that answered NXDOMAIN (Non-Existent Domain) after the scan, so a dead name here is one that exists and is dead. The census counts 2,887,602 of them. Earlier editions counted dead names a different way, so the rates on this page can’t be set beside theirs.
The rate gap
Across new gTLDs, meaning gTLDs other than com, net and org, 393,003 of 65,234,684 names are dead, a dead rate of 0.602%. The country-code domains sit at 2.028%, 2,096,902 of 103,380,131. The three legacy endings are lowest at 0.181%, 346,527 of 191,853,570. On the same measurement, the country-code rate is about 3.4 times the new-gTLD rate.
Three endings drive that. .tk has 948,088 dead names, 79.31% of its 1,195,397. .cf has 336,712 (78.67% of 428,015) and .gq 275,341 (76.20% of 361,344). Together they hold 74.40% of the country-code dead names. Leave them out and the other country codes sit at 0.529% (536,761 of 101,395,375), a little under the new-gTLD rate.
“Dead” is the census engine’s definition: the scan asked for the domain’s name servers, got a definitive empty answer, and no address or mail record resolved either. A dead name is still in the count, because it did not answer NXDOMAIN. It is a separate group from the dark matter, the delegated names the scanner could not grade. A dead domain publishes nothing, the census grades it F, and it is counted inside the 332,492,319 domains carrying a grade.
The new gTLD group also holds the older generic endings such as .info, .biz and .pro, which were on sale before that wave began. Take the 15 of those in the file out and the rate for the remaining 335 endings falls slightly, to 0.548% (308,070 of 56,242,264). Much of the difference is .gov, which has 71,040 dead names out of 87,922. Nearly all of those dead names are missing from CISA’s registry of .gov domains but still sit in the source lists the census reads, as the .gov piece explains.
Where the dead names sit
The averages hide where the dead names sit: a small set of cheap, promotion-driven endings. Among new gTLDs with at least 100,000 names, these have the highest dead rates:
| ending | names in the census | dead | dead rate |
|---|---|---|---|
| .cfd | 615,530 | 64,250 | 10.44% |
| .bond | 1,176,122 | 58,682 | 4.99% |
| .icu | 552,122 | 22,557 | 4.09% |
| .live | 619,478 | 8,801 | 1.42% |
| .vip | 1,962,663 | 18,345 | 0.93% |
| .top | 6,291,944 | 39,915 | 0.63% |
| .shop | 4,316,695 | 22,472 | 0.52% |
About one in ten .cfd names in the census no longer answers.
The dead names are concentrated as well as frequent. Ten endings, led by .gov, .cfd, .bond and .top, hold 320,512 of the group’s 393,003 dead names, 81.55%. Of the 64 new gTLDs with at least 100,000 names, 4 have a dead rate of 1% or more and 3 have 2% or more. The other 60 sit under 1%.
The country-code and legacy side
Of the 74 ccTLDs with at least 100,000 names, 11 have a dead rate of 1% or more and 8 have 2% or more. .tk, .cf and .gq sit far above the rest. Next come .ae at 22.79% (48,537 of 212,941), .bd at 5.45% (5,577 of 102,360) and .tw at 3.63% (11,376 of 313,531). .ua is at 2.64% (9,724 of 367,755) and .at at 1.20% (12,226 of 1,021,601). The census records the rates and does not record why.
The legacy endings are close to one another. .com has 300,544 dead names, 0.18% of 167,247,780. .net sits at 0.20% (24,598 of 12,417,769) and .org at 0.18% (21,385 of 12,188,021). Only .tk and .cf have more dead names than .com, and .com’s count comes from its size: its rate is below the new-gTLD average.
Unreachable and indeterminate names
Dead is the definite end. Names that answer badly land in the unreachable or indeterminate states instead, and those are counted separately. .garden has 10 dead names out of 209,424, and 30,093 (14.37%) the scanner could not reach at all. An ending can look healthy on the dead column and still carry a large share of names nobody could grade. The dark matter piece covers those two states.
How the rates were computed
Every figure on this page comes from the per-TLD rows of the census file. For each ending, dead is the inventory minus the names the file marks alive, and the dead rate is dead over inventory, the names in the census for that ending that exist. The file publishes 520 of the 1,434 TLDs the census reached, covering 360,468,385 of the 361,301,790 names that exist (99.77%). The group totals above sum those 520 rows.
The ccTLD and gTLD grouping follows the IANA (Internet Assigned Numbers Authority) delegation type, so every two-letter ending and every internationalised country-code ending counts as a ccTLD. That puts .co, .io, .ai, .me and .tv with the country codes, whatever they are marketed as. The tables use a floor of 100,000 names because in a small ending a few hundred names move the rate by whole points. Smaller endings with higher rates exist in the data, and the census data page links the per-TLD file, which has a dead_pct column for each of its 520 endings, so you can apply your own floor.
What it does not mean
A dead domain is an abandoned one, which is a different finding from an insecure one. This page counts resolution, and grades play no part in it. A .cfd domain that still resolves can score anywhere; the point here is only that a large share of the ending never resolves at all. If you are choosing an ending, a high ghost-town rate is a signal about the neighbourhood’s churn, and it says nothing on its own about whether the live domains are configured well. The how the domains were counted page explains how alive, dead and graded are separated. For grade shares by ending, split by country-code and generic, the TLD statistics page and the explorer carry them.
See your own
The free scan reads your domain live. If it resolves, you get a grade; if your ending is a ghost town, that has no bearing on your own result. If the scan finds no name servers for a domain you own, check the delegation at your registrar first; the name server fix page covers what to look for.
Figures from the defaults.exposed census, data as of 12 October 2026. Census numbers move every month; the current values are on the data page.