A 200 status is not proof a link is alive.

A quarter of popular links from 2014 no longer work. 25% of the 2014 links we could check were broken in September 2026, against 17% from 2019 and 5% from 2024. Most of them fail the honest way, with an error. But some fail in a way that link checkers miss: the link still returns a normal “200 OK”, the signal that a page loaded fine, while the domain behind it now belongs to someone else.

Of the 400 links from 2014 that still answered with a 200, 10 led to a gambling site, a redirect into an unrelated site, or a “page not found” dressed up as a working page. That is about one in 40. Across all three years we found 16 domains that had been registered again after the story was posted, and were still answering.

Link rot by year

Links decay steadily rather than all at once, which is the pattern the link rot act on this site describes as a half-life. Roughly one in twenty links had broken within two years, one in six within seven, and one in four within twelve.

What “dead” looked like

Of the 227 links that returned an outright failure:

• 99 returned “not found” (404 or 410): the site survived, the page did not.
• 53 pointed at domains that no longer exist at all.
• 39 pointed at servers that never answered.
• 15 failed on an expired or broken security certificate.
• 21 failed with other server errors.

These are the easy cases. A browser shows an error, a link checker flags it, and nobody is misled.

When a working link lies

The harder cases answer perfectly. We found 8 links from old blogs and open source projects whose domains now serve something else:

• toonormal.com, posted in 2014 as “4k, HDMI, and Deep Color”, now serves an online gambling page.
• thecodist.com, posted in 2014 as “Your Progress As A Programmer Is All Up To You”, now serves an online gambling page.
• haste-lang.org, posted in 2014 as “Haste language”, now serves an online gambling page.
• howtotex.com, posted in 2014 as “Essential LaTeX packages”, now redirects, along with other expired domains, to an unrelated site that shows only a loading screen.
• vindinium.org, posted in 2014 as “Vindinium: An artificial intelligence programming challenge”, now serves an online gambling page.
• otacoo.com, posted in 2014 as “Postgres 9.4 feature highlight: Indexing JSON data with jsonb data type”, now redirects, along with other expired domains, to an unrelated site that shows only a loading screen.
• libregraphicsworld.org, posted in 2019 as “Olive, a new non-linear video editor”, now serves an online gambling page.
• kynosarges.org, posted in 2019 as “Samurai Helmets”, now serves an online gambling page.

None of these are linked here, on purpose. Every one of them still returns 200. To a link checker, every one of them is fine.

We also found links where the old address now quietly forwards to a site’s homepage, or to a page that says “not found” while telling the browser everything is normal. Those look alive and deliver nothing.

Domains that came back

We looked up when each domain was last registered. 16 domains from the 2014 and 2019 samples had been registered after the story was posted, which means they lapsed and someone registered them again, and they were still answering in 2026. Some now host an unrelated business. Some show a holding page. And several show what looks like the original site again.

That last group is the strangest. We cannot tell from the outside whether the original owner came back, or whether someone bought the lapsed domain and restored its old pages from an archive. The second is a known tactic, because a domain that other sites still link to carries search ranking, and a convincing copy of the old site keeps it. Either way, the page a reader finds today is not necessarily the one the link was written for.

How this compares

Pew Research found in 2024 that 38% of web pages that existed in 2013 could no longer be reached. Our figure for 2014 is lower, around a quarter, which makes sense: links that reach the front page of Hacker News point at pages people cared about, on sites more likely to be maintained. A 2014 Harvard study found that about half of the links in US Supreme Court opinions no longer pointed to the original material.

Most published figures, including Pew’s, count a link as alive if it loads without an error. Unlike those studies, we also looked at what the working links now show. The disguised failures are a small share of all links, but they are the ones that cannot be caught by the usual checks, and the share grows with age.

What to do about it

1. Link to a saved copy as well as the original. Save the page to the Internet Archive’s Wayback Machine when you link to it, and put that link next to yours. Of the broken 2014 links we looked up, 73 of 75 already had a saved copy. The page usually survives somewhere. Readers just can’t reach it from the link.
2. Check links for meaning, not just errors. A link checker that only reads status codes will pass every hijacked link in this study. Compare page titles against what you expected.
3. Look again at links to small sites. Personal blogs and side projects are the most likely to lapse, and lapsed domains are the ones that get reused.
4. Redirect your own old addresses when you move or rename pages, so the links other people made to you keep working.

Method and limits

We sampled 50 stories a month, at random, from Hacker News stories with more than 100 points in 2014, 2019 and 2024: 600 per year, 1,800 in all. On 23 September 2026 we requested each link once, following redirects, with a browser-like user agent so that sites filtering obvious bots would respond normally. Links that came back “forbidden” or rate limited were counted as blocked and left out, because they may well work for a person. That left 1,557 links. A link was counted as hijacked if its page title now advertised gambling or adult content unrelated to the story, if it showed a domain-parking page, or if several unrelated old domains now redirected to the same new site. Redirects to a site’s homepage and “not found” pages served as normal were counted separately. Registration dates came from public RDAP records. Archive copies came from the Internet Archive’s capture index, for 75 of the 132 broken 2014 links, because each lookup is slow. The rules are simple and will miss some cases, so the disguised figures are best read as a floor. The full data is linked below.