Broken link checker

Outbound links decay whatever you do. Companies close, publications reorganise their URLs, documentation moves to a new version and drops the old one, and personal sites disappear when a renewal is missed. An article that was carefully sourced when it was written accumulates dead references at a rate nobody controls, and the older and better the article, the more it has.

That makes this maintenance rather than a task with an end. A pass every few months over the pages that actually earn traffic is enough, and it pairs naturally with the freshness review you would be doing on those pages anyway. The pages nobody visits can wait, since the cost of a dead link is paid by readers and there are none.

Internal links are the ones to fix immediately, because they are entirely yours. They break in bulk at a specific moment, a permalink change or a migration, and they strand any page that was only reachable through them. If the sitemap still lists the old URLs as well, the crawl spends its budget confirming that a set of pages you deleted are still deleted.

First look for the resource at its new home, since most of the time the page moved rather than vanished. Searching the exact title usually finds it in one attempt, and linking the current location is better for the reader than anything clever.

When it is genuinely gone, an archive snapshot preserves the reference and lets a reader check the claim you attached to it. That matters more than it looks: silently deleting the link leaves a sentence asserting something with the evidence removed, which is a worse outcome than the broken link was.

If neither is available, edit the claim rather than the link. A statement whose only support has disappeared should be softened, attributed differently, or cut. And when you fix internal links, update them in the content rather than relying on the redirect to keep working, because a chain of redirects accumulated over three migrations eventually ends somewhere nobody intended, usually at a canonical that no longer matches.

A typical result on an eight-year-old blog: forty links sampled, seven failed, and the seven are not one problem but four. Two internal 404s, three dead outbound links, one timeout, and one redirect chain that does eventually land on a 200. Each class gets a different treatment, which is why the report shows status codes rather than a count.

The two internal 404s turn out to share a cause: a permalink change in 2021. The navigation was updated at the time, but two links written inside article bodies still point at the old addresses, which is the classic split, menus get maintained and prose does not. The fix is a redirect from each old URL plus editing the two sentences to link the current address, and a status check to confirm the old paths now answer honestly rather than soft-404ing.

The three outbound deaths each get their own answer. A free tool that shut down: the sentence keeps its claim and the link moves to an archive.org snapshot. A documentation page that moved during a rebrand: thirty seconds of searching the exact title finds the new home. A news article now returning 410: the paragraph leaned on it as evidence, so the claim gets softened rather than left standing unsupported. The timeout is the row that needs judgement rather than action: the destination loads fine in a browser and simply refuses automated clients, which is worth knowing and not worth fixing.

The redirect chain, http to https to www to 200, works for visitors and wastes hops for everything else, so the href gets updated to the final URL. Then the schedule: a pass like this quarterly on the pages that earn traffic, always after a restructure, and a glance at the sitemap at the same time, since the restructure that broke the links usually left stale URLs there too. The technical audit bundles all three so the pass stays one job.

The longer game is linking in a way that ages well, because the cheapest broken link is the one never created. Link the stable page rather than the deep one: a project's documentation home outlives any individual doc URL, an organisation's page outlives this quarter's campaign page. Prefer the original source over a blog quoting it, since the mirror dies first. And when a claim depends on something time-bound, a price, a limit, a version, date it in the sentence, so that when the link eventually dies the text still tells the reader what was true and when. None of that prevents rot; all of it lowers the cost of each break when it comes.