Linkbruno

Research · 3 November 2026

Link rot on Hacker News: how many links from 2015 to 2025 still work

Dead links, by the year the story was posted

2015: 19%, 2016: 17%, 2017: 14%, 2018: 12%, 2019: 10%, 2020: 10%, 2021: 10%, 2022: 9%, 2023: 5%, 2024: 4%, 2025: 4%

We took 11,000 popular Hacker News stories, 1,000 from each year from 2015 to 2025, and checked every link with Linkbruno’s checker. 19% of the links from 2015 are dead now: the page is gone, the domain has lapsed, or the link sends readers somewhere else. The newer the story, the lower the share, down to 4% for links posted in 2025.

Another 16% of all the links didn’t let a link checker in. Bot protection from DataDome, Cloudflare and others answered instead of the site. A checker that goes by status codes reports those as broken, though they say nothing about the page behind them.

Links checked
10,962
Dead, all years
10%
Couldn’t check
16%
Still work
74%

Dead links by the year the story was posted

Share of the 1,000 links from each year. The figure on the right is the dead share. Hover a row, or tab to it, for the rest.

  • Dead
  • Couldn’t check
  • Still works
2015
19%
2016
17%
2017
14%
2018
12%
2019
10%
2020
10%
2021
10%
2022
9%
2023
5%
2024
4%
2025
4%
Show the numbers as a table
YearLinksDeadCouldn’t checkWorks, same addressWorks, moved
20151,000190 (19%)164 (16%)470176
20161,000169 (17%)189 (19%)466176
20171,000145 (14%)166 (17%)505184
20181,000119 (12%)194 (19%)520167
20191,00098 (10%)223 (22%)517162
20201,000103 (10%)160 (16%)596141
20211,00095 (10%)148 (15%)620137
20221,00087 (9%)116 (12%)663134
20231,00052 (5%)142 (14%)695111
20241,00040 (4%)131 (13%)74386
20251,00039 (4%)120 (12%)77962

The 1,137 dead links, by what a reader gets when they click.

Page not found (404 or 410)
48%
Domain no longer exists
17%
Server doesn’t answer
16%
Redirects to the home page
6%
Server error (5xx)
5%
Certificate error
5%
Parked or for-sale domain
1%
Redirects in a loop
1%

A 404 is the easy case: any link checker reports it. The others are the ones that get missed. A domain that no longer exists has no server to answer, and at least one free checker lists that as “unverified” rather than broken. A redirect to the home page answers 200, which a status-code check reads as fine, and so does a domain that lapsed and became a parking page, with ads where the article was.

16% of links turned a link checker away

The 1,753 links we couldn’t check, by who answered instead of the site.

Cloudflare
36%
DataDome
26%
Took the connection, never answered
11%
HUMAN (PerimeterX)
10%
Refused (403 or 406), no vendor named
5%
Rate limited (429)
5%
Sent us through a sign-in step
4%
Akamai
2%
AWS WAF
1%
Payment required (402)
1%
Imperva
0%
Unavailable in Europe (451)
0%

These answers come from the service in front of the site, and they say so: a DataDome or Cloudflare challenge page, a vendor header, LinkedIn’s made-up 999. Most come with a 403, a 429 or a 503, so a checker that reads the status code lists them as broken, and the reader learns to ignore its reports. Why link checkers see 403 and 429 explains the mechanics.

We counted these as unchecked, not dead and not working. Some of the pages behind them are likely gone too, so the dead shares above are a floor. We didn’t try to get past the walls.

The most upvoted stories in the sample whose links are dead

  1. Why isn't the internet more fun and weird?
    2019 · 1,516 points · Page not found (404 or 410)
    https://jarredsumner.com/codeblog/
  2. “Swift will be open source later this year”
    2015 · 1,334 points · Page not found (404 or 410)
    http://www.apple.com/live/2015-june-event/9d2ad033-d197-4009-96a7-2a97fd044cb7/
  3. GitLab Database Incident – Live Report
    2017 · 1,162 points · Page not found (404 or 410)
    https://docs.google.com/document/d/1GCK53YDcBWQveod9kfzW-VCxIABGiryG7_z_6jHdVik/pub
  4. Google Drive flags file only containing “1” for copyright infringement
    2022 · 1,117 points · Page not found (404 or 410)
    https://twitter.com/emilyldolson/status/1485434187968614411
  5. Apple Silicon M1: Black Magic Fuckery
    2020 · 1,042 points · Certificate error
    https://www.singhkays.com/blog/apple-silicon-m1-black-magic/
  6. Nine things I learned in ninety years
    2025 · 1,006 points · Page not found (404 or 410)
    http://edwardpackard.com/wp-content/uploads/2025/09/Nine-Things-I-Learned-in-Ninety-Years.pdf
  7. Judge in Google case disturbed that 'incognito' users are tracked
    2021 · 927 points · Redirects to the home page
    https://www.bnnbloomberg.ca/judge-in-google-case-disturbed-that-even-incognito-users-are-tracked-1.1569065
  8. Play Counter-Strike 1.6 in your browser
    2020 · 878 points · Domain no longer exists
    http://cs-online.club/
  9. Pornhub Bypasses Ad Blockers with WebSockets
    2016 · 873 points · Page not found (404 or 410)
    http://blog.bugreplay.com/post/152579164219/pornhubdodgesadblockersusingwebsockets
  10. Reddit's favorite products in one place
    2022 · 839 points · Page not found (404 or 410)
    https://looria.com/reddit/overview

Each title opens the Hacker News discussion, which is still there.

96% of dead links have an archived copy

We looked up 70 of the dead links, picked at random, in the Wayback Machine. 67 of them had a good copy saved no later than 90 days after the story, which is the page as Hacker News readers saw it. That’s a small sample, because archive.org asked us to slow down after it, but it’s enough to say that most do. Linking to that copy is often the best fix for a dead source. The Wayback Machine checker finds one for any address, and how to fix broken links covers when a copy is the right fix.

How we did it

  • The stories. Every Hacker News story from 2015 to 2025 with 100 or more points and a link, from the HN Search API: 153,401 stories. Links back to Hacker News itself were left out. We drew 1,000 at random from each year, with a fixed seed so the draw can be repeated. With 1,000 a year, each year’s share is within about 3 percentage points of the share for all of that year’s stories, 19 times in 20.
  • The check. Each link went through the same checker Linkbruno runs for its customers, on 3 November 2026, from one connection in Europe. It follows redirects one hop at a time, asks again with a GET when a HEAD answer looks wrong, and reads the page to spot “not found” pages and parking pages. It asked each site at most once a second, as our crawler, with its own user agent.
  • Second look. 3,034 links didn’t work or couldn’t be checked the first time. We checked those again 7 to 30 minutes later, and the second answer is the one counted. 148 of them worked then. An outage longer than that counts as dead here.
  • Dead means a reader doesn’t get the page: a 404 or 410, another error that stayed on the second look, a domain that doesn’t resolve, a server that never answers, a certificate error, a page that says it isn’t there, a parking page, or a redirect from an article to a site’s home page. We read 25 of those home page redirects by hand: 20 had lost the page, 3 were projects that had moved to a site of their own, and 2 were hard to call. After that a move like those three counts as working.
  • Couldn’t check means something other than the page answered: bot protection, a rate limit, a sign-in step the site sends every visitor through, a paywall that answers 402, or a site that refuses European visitors with 451.
  • Still works includes pages that moved and redirect to their new address, and pages behind a paywall or a cookie banner. It doesn’t mean the content is the same: we didn’t compare pages with what was posted, so links that drifted to different content count as working.

What it found wrong with our own checker

We read the odd results by hand before publishing, and four of them were mistakes in Linkbruno’s checks, not in the links. Each is fixed for customers too, and the links were checked again with the fix:

  • A site whose secure connection failed, when the certificate itself looked fine, came out as working. Chrome couldn’t open the one we tried either. It’s now reported.
  • time.com answered 406 to our checker, to curl with a browser’s user agent and to headless Chrome. We called that dead. A 406 is now treated like a 403: probably bot protection, open it to check.
  • Our parking-page check matched the name of a parking service, dan.com, inside longer names like belkadan.com. A third of the “parked” domains were real blogs that link to one. It now matches whole names only.
  • Sites like nature.com send visitors to a sign-in server and back to see if they’re signed in. A browser gets the page; we went round in circles and called it a redirect loop. That’s now reported as blocked, not broken.

And one it still gets wrong. It found no page that answers 200 while saying it isn’t there, a soft 404, but we found two by hand: a Lockheed Martin job page titled “404”, and a DreamHost “Site not found” page. Both count as working here, so the dead shares are a little low.

Use the data

Every link, with its story, its points, what we found and when, is in one CSV. You’re welcome to use it and the charts. Please link to this page when you do.

Download the CSV (11,000 rows)

The links on your site age the same way. Link rot covers what other studies found and a routine for keeping links working, and how to check external links covers the dead outbound links this study is about. To check one page now, use the page link checker, or the free scan for 25 pages of a site.