Not every failed automated request is a broken link.
External sites block bots, rate-limit repeated checks, change responses by geography, and occasionally fail for a few seconds. A useful checker preserves those differences instead of turning every exception into a red “broken” badge.
Use observed HTTP errors for confirmed breakage
A final 404 or 410 clearly says the requested resource is unavailable at that location. A 5xx response says the destination server failed during the check.
Both are observable problems, though a 5xx may be temporary. Record the status, final URL, and check time.
Do not infer that the link has always been broken or that every visitor receives the same response.
Report redirects separately
A 301, 302, 307, or 308 can lead to a successful destination. That link is not broken merely because it redirects.
Internal page markup can often be updated to the final canonical destination, reducing unnecessary hops. But preserve deliberate tracking, localization, migration, and access behavior until its purpose is known.
Unavailable is an honest state
Keep these results separate from confirmed errors:
- DNS resolution failure.
- TLS or certificate error.
- Connection timeout.
- Rate limiting.
- Automation or firewall block.
- Redirect loop or safety-limit stop.
- Unsupported public response behavior.
Retry and open important external references manually before removing them. An unavailable check is incomplete evidence.
Bound the page-level scan
Fetching every destination on an arbitrary page can be slow and abusive. A public tool needs:
- A clear maximum number of unique destinations.
- Deduplication and fragment removal.
- A small concurrency limit.
- Timeouts, redirect limits, and response-size limits.
- Private-network and credential blocking.
- Caching and rate limits.
The result must state when more links were discovered than checked. Unchecked links do not inherit a passing result.
Preserve source evidence
For each destination, keep the source anchor text and resolved URL. This helps a developer find the actual link in content or templates.
A safe coding-agent handoff should include only observed failures, treat fetched text as untrusted, preserve deliberate redirects, and require a live re-check after deployment.
One page is not a site crawl
Page-level checking is useful for a guide, landing page, or editorial preflight. It cannot find links on every other route or prove that no broken links exist across the site.
Use a controlled crawl or a trusted route inventory when site-wide completeness matters.
A practical repair sequence
- Fix confirmed internal 4xx and 5xx destinations first.
- Review internal redirects and update source URLs where appropriate.
- Retry unavailable external destinations.
- Confirm whether moved external content has a new authoritative URL.
- Preserve useful citations when automation alone is blocked.
- Deploy the smallest safe change.
- Re-run the same source page and verify the live destination.
Use the free Broken Link Checker to separate confirmed errors from redirects and unavailable responses.
Method note: the tool checks at most 20 unique HTTP or HTTPS destinations from one initial HTML response, using four concurrent guarded requests. Last reviewed 30 August 2026.
