But the functionality feels like it'd be prone to breakage as new text is added to the document (possible in Wikipedia articles for example). But it'd be super-helpful for search engines as they will have a relatively recent copy of the text, and can auto-generate the links on the backend before showing the search results to the user.
You are right that the link is somewhat time limited. However, tools like emacs and ctags show this can be somewhat worked with. Instead of the very specific links that grep creates, generate one that has a location, but also a small pattern that indicates it. To that end, you could generate a link to the word "breakage" with a selector saying where it is expected. And the browser can then just fudge around that selector to find that word. Or highlight where it was expected.
Yes, it would fail if the target text no longer exists in the page but link rot is already commonplace. It would be important for it to ignore DOM elements, it can't fail because the target text crosses two paragraphs or a span.