Broken link

An internal broken link points within your own site. It is unambiguous, entirely within your control, and always worth fixing.

An external one points at somebody else's site. It matters, but it produces most of the false alarms, because other servers block automated clients while serving browsers normally.

The distinction determines both the priority and the confidence you can place in an automated report.

404 Not Found — broken. No representation for that URL.

410 Gone — broken, and deliberately so.

5xx — the server failed. Often transient; re-test before editing anything.

401 / 403 — the target exists and declined the request. On an external link this is usually bot protection, not absence.

429 — rate limited. The checker asked too often, not a defect in the link.

200 — reachable. Says nothing about whether it is the intended page.

A soft 404: the target says it does not exist while returning 200. A link checker reports it as working.

A redirect chain ending in a 404: the first hop looks fine, the destination does not exist.

A broken image or script: the same defect with a different element. A missing image is a 404 the visitor perceives as a gap rather than an error page.

`curl -I -L <url>` issues a HEAD request, follows redirects and prints each hop's status. Some servers reject HEAD; `curl -sS -o /dev/null -w '%{http_code}' <url>` falls back to a GET.

At site scale this requires the list of URLs first, from a crawl or the sitemap — which is the part that does not scale by hand.

External links rot on their own schedule. Studies of link persistence consistently find that a substantial share of cited URLs stop resolving within a few years, without anybody on either side doing anything wrong. An article written in 2019 has simply lost some of its references.

Internal links break in bursts rather than gradually. A migration, a redesign, a bulk category rename — each produces a batch of them on a single day, and none between.

Content deletion produces them silently. Unpublishing an article removes the page; nothing removes the links from the six other articles that referenced it.

Typos in hand-written HTML and Markdown persist indefinitely, because the author sees the rendered link and never clicks it.

The two patterns need different responses: external decay wants a periodic re-check, internal breakage wants a check tied to the change that caused it.