Find the broken links that point away from your site
Crawl Audit can now check the links from your pages to other sites, off by default and bounded, and reports the ones that return 404 or 410 or whose host no longer exists, with the pages that link to them and the search-bot traffic those pages get.
Every Salience crawl now records where your pages link out: the external URLs, how many of your pages link to each, and a sample of the linking pages with the anchor text and where the link sits. The crawl audit's coverage summary lists your ten most-linked destinations, so you can see at a glance where your site sends people.
Switch on Check external links for a site, or pass check_external_links when starting a crawl, and the crawler also checks those links. It is off by default because it is the only part of a crawl that contacts servers you do not control, and it is deliberately bounded and polite: at most 200 URLs per crawl chosen by how many of your pages link to them, one HEAD request each with no body downloaded and no redirect followed, each host's robots.txt read and honoured first, requests to one host at least two seconds apart, and no further requests to a host that asks us to slow down. Payment and sign-in hosts are never contacted. The requests identify themselves as SalienceBot from the crawler's fixed address.
The audit then lists Broken external links: links that returned 404 or 410, or whose host could not be reached, each with the pages that link to it and the verified search-bot and human traffic on those pages, so the links readers and bots actually meet come first. External links that redirect are listed separately with the destination host, so you can point the link at its final URL; a moved link is worth knowing about but is not broken. Everything else is honest about what it is: a link that answered 401, 403, 429, a server error or a timeout could not be checked and is counted as unavailable with its reason, never reported as broken, and links beyond the budget or disallowed by the other site's robots.txt are counted as unchecked. Zero broken links is never taken as proof that every link works.
Only links in the original HTML are seen; links added by JavaScript, links inside PDFs and scripts, and the content of the linked pages are outside this check. Start a new Salience crawl to collect the evidence; crawls captured before this release show the check as unavailable. The public API, MCP get_crawl_audit and salience crawl-audit --json carry the findings and coverage.
See Crawl Audit help for what broken, unavailable and unchecked mean, the request budget and the politeness rules.