What is survivorship bias?

The classic version of this involves studying successful companies to find what makes companies succeed, without checking whether the failed companies did the exact same things. In AI visibility, the version that shows up is studying citations you received and reverse-engineering what "worked," without ever looking at the queries or topics where your content was in the mix but didn't get cited, or wasn't retrieved at all.

It also shows up at the crawl level: a domain that couldn't be reached, whether from a firewall, a timeout, or a broken certificate, simply drops out of the dataset. Describe the remaining domains as "the internet" and you've quietly restricted your conclusions to the domains that survived being crawlable, which is not the same population as all domains.

Selection bias and survivorship bias are close cousins, selection bias is about how the group got assembled in the first place, survivorship bias specifically about looking only at what made it through a process and forgetting to account for what didn't.