DOM is not very "shifting" in projects like these... The longer the project lives, the more tests it accumulates. A "good" practice is to rely on element ids for automation (i.e. clicking, finding form inputs etc), but often times it's not possible / hard to know in advance what the id will be, especially with all the HTML generation going on... so, tests would often use XPath to find where to click.

And nobody really wants to change those tests... I mean, yeah, there will be changes ever now and then, but not a lot and not enough to render out-of-date scrappers completely useless.