Implementing Into<X> vs having an into_X function

Yes, I've read the docs about Into and From. I understand their use cases and how they differ, but it still remains unclear to me when should a type T implement From<S> versus having a function like fn into_t(self) -> T.

I know that implementing From<S> allows me to have code like t: T = s.into() instead of t: T = s.into_t(), but are there any guidelines where one shouldn't "abuse" of the From trait? I've tried referring to examples from the standard library but I've found mixed results...