I think this is a classic case where your research goals should determine your choice of a method. In particular, how are you relying on an interpretive search for meaning, or a pattern produced by counting codes.
Instead of the language of Hsieh & Shannon, I prefer to think of the first two approaches as inductive (i.e., creating the codes through reading the data) and deductive (.e., using a codebook derived from some external source). Of course, there is also a so-called "hybrid" approach that begins with a core of deductively determined codes and then adds to that by working with the data inductively .
I think either of these two is quite different from counting pre-determinded codes through word search and Keyword-in-Context (KWIC) approaches. In particular, I would question the extent to which this latter option is indeed naturalistic, because it typically stops with a counting of codes, rather than searching for interpretation and meaning in the data itself. As such, it is mostly a way to transform qualitative data into quantitative analyses.