I did something similar in my paper "Meno II: A self-referential Socratic dialogue about memory and computer programming" [1] (published in The Reasoner). Every length-200 block of text is carefully chosen so as to steganographically encode one character of a computer program, and Socrates and Meno decode the program in real time by examining the words they've said.

I didn't do it entirely by hand, but I didn't just one-shot it with AI either (good luck). Instead, I wrote the whole dialogue in a special format specifying alternate words [like|similar to] [this|what I'm [currently doing here|doing right now]]. For each length-200 block, the number of variations grows exponentially in the number of "options" to thus choose from, so it becomes feasible to have a computer systematically choose options so as to hide the desired secrets.

The same method could be used for monospace alignment, though I suspect it would be harder the narrower the page be.

[1] http://philpapers.org/archive/ALEMIA-4.pdf