The Most-Edited Answer Isn't the Current One
Search a specific error string from a library you're using and the top result is almost never the current documentation. It's a Stack Overflow thread from six years ago, four hundred upvotes, an accepted answer with three edit notices stacked underneath it, and a comment section arguing about whether the fix still applies. It usually doesn't. The API moved on. Nobody went back and struck the answer through.
The ranking isn't wrong, exactly. It's measuring something real: this page got linked to, argued over, corrected, revisited. That's a genuine signal of engagement. The problem is engagement and correctness are different axes, and the ranking only has instruments for the first one. A page earns authority by accumulating attention, and attention keeps accumulating on old pages because they've had longer to collect it and because new pages start every search from zero.
Compare that to a vendor's current docs page for the same API. It has no upvotes, because there's no voting mechanism. It has no edit history visible to the crawler, even though someone updated it last month, because the CMS doesn't expose diffs the way a wiki does. It has fewer inbound links than the Stack Overflow thread, because nobody links to docs the way they link to the answer that solved their specific problem at 2am three years ago. The current page is right and it looks, to a ranking algorithm, like a new site with nothing behind it.
So the incentive facing anyone who wants their answer found isn't "be correct," it's "be the place people already go." Once a Stack Overflow answer has enough votes and inbound links, it's largely immune to being wrong later. The correction, if it comes, arrives as a comment forty replies deep, itself unranked, itself invisible to anyone who stops reading after the accepted answer's code block. The page keeps winning searches for the version of the problem that no longer exists, because it already won the search for the version of the problem that used to exist, and ranking is mostly a function of what already won.
This isn't limited to Stack Overflow. Any format that rewards accretion, wikis, FAQ pages, forum threads, GitHub issues with fifty resolved-but-reopened comments, will outrank a clean replacement written after the accretion started. The old page has more stuff attached to it, and more stuff attached looks like more authority, even when the stuff attached is disagreement, deprecation notices, or a decade of drift away from whatever's true today. A page that says "this is current as of right now" structurally cannot compete with a page that says "here is everything anyone has ever said about this," because the second one has more surface area for links to land on.
The fix people reach for is "just check the date," which assumes the reader notices there's a date, checks it against the current version number, and holds both in their head while reading code. That's asking a lot of someone who searched an error message because they wanted the error to go away, not because they wanted to do software archaeology. The date is metadata. The rank is the actual argument being made, and the rank says: this is the one everyone else already trusted.
None of this is a plot. Nobody designed search to prefer the fossil over the living tissue. It just turns out that "has been touched a lot" is much easier to measure than "is currently true," and systems build what they can measure. The dead API version keeps outranking the docs that replaced it because the dead version got there first and never stopped collecting evidence of its own popularity, while the truth had to start over every time it changed.