Link rot is measurable and boring, which is why it goes unfixed. The question is not whether broken links are bad; it is whether the time you spend running and triaging a checker returns more than it costs. Here are the numbers that decide it.
Three separate costs, and only one of them is SEO. First, crawl waste: every 404 an engine follows is a fetch it does not spend on a page you care about. Second, dead outbound links on a monetised page are lost clicks — an affiliate link that 404s converts at exactly zero. Third, support load: a broken download or docs link generates tickets. On a 5,000-page content site, a typical first crawl surfaces somewhere between 40 and 300 broken links, most of them clustered in old posts pointing at products and tools that have since moved.
lychee --verbose --no-progress --accept 200,206,429 './**/*.html' will crawl a static site's output directory in place, no server needed. Handles concurrency well and has a GitHub Actions wrapper.linkchecker --check-extern -o csv https://example.com > report.csv.wget --spider -r -nd -nv -l 5 -o crawl.log https://example.com, then grep the log for "broken link".Screaming Frog's licence is around £199 per year and removes the 500-URL cap, adds scheduling and JavaScript rendering. Sitebulb and Ahrefs Site Audit sit higher, in the $100+ per month range, and what you are buying is not link detection — it is history, alerting and the ability to hand a report to someone who does not use a terminal. If you own one site under 10,000 pages, the free tools cover it. If you audit client sites for money, the licence pays for itself the first time scheduling saves you a manual crawl.
lychee on a static export with 20 concurrent requests handles roughly 5,000 internal links in two to four minutes; external links are far slower because you are at the mercy of remote servers and rate limits. Expect 15-40 minutes for a full external check of a mid-sized site, and expect a stack of false positives from sites that return 403 to non-browser user agents. Budget more triage time than crawl time.
Triage by status code, not alphabetically. 404 and 410 on internal links go first and are always your fault. 301 chains longer than one hop are worth flattening. 403 and 999 are usually bot blocking, not breakage — spot-check two in a browser and then ignore the class. 429 means you crawled too hard; re-run that host with --max-concurrency 4. Connection timeouts need a second run before you believe them.
The real return comes from running it on a schedule and only looking at the delta. A weekly CI job that fails the build on new internal 404s costs a few minutes of compute and stops rot at the source. A one-off audit that finds 200 broken links and produces a spreadsheet nobody opens costs a day and returns nothing.