This interpretation is error-prone. (seq 1), for instance, throws an IllegalArgumentException at runtime. Meanwhile, (seq nil) returns a falsey value even though in many cases it's possible to transparently pass nil in place of a collection of any kind.
Essentially, seq already assumes that its argument is seqable. Don't use it to test an argument of unknown type for seqability; prefer something like the seqable? function from core.incubator [1], or a different core function like coll? or sequential? where the semantics of one of these will do.
[1] https://github.com/clojure/core.incubator/blob/master/src/ma...
Of course, not-empty is implemented with (when (seq)), so there's no REAL difference. At the end of the day, (when (seq)) is what people use.