I think some of the confusion here arises because the term “suffix trie” means more than just “a trie that holds suffixes.” Rather, a suffix trie typically means “a prefix trie holding all the suffixes of a given string.” This contrasts with “prefix trie,” which typically stores an arbitrary collection of strings rather than all prefixes of a given string.
The reason suffix trees are useful is the following fact, sometimes called the fundamental theorem of stringology:
A string x is a substring of a string w if and only if x is a prefix of a suffix of w.
For example, “irate” is a substring of “pirates” because it’s a prefix of the suffix “irates.”
This fact is why suffix tries are so good at substring searching. Suppose you want to see whether x is a substring of w. And further suppose that, somehow, you obtained a suffix tree for w. Then you can just walk the suffix tree from the root downward and see whether you can read x without falling off the tree. If so, x is a prefix of some suffix of w, so x is a substring of w. If not, x isn’t a prefix of any suffix of w, and so x isn’t a substring of w.
As @Ed Staub’s answer shows, you could just as easily do this by using a trie that stores all the prefixes of w in reverse, then checking if x is a suffix of any prefix of w. But in practice it’s easier to think about holding all the suffixes in a prefix trie and so that’s what we do.