Simulated annealing and similar techniques such as Tabu Search are good for problems that have large search spaces due to combinatorial explosion (https://en.wikipedia.org/wiki/Combinatorial_explosion) and can be expressed as a state and an associated cost/fitness function.
This is true provided your cost function is "continuous" w.r.t. distance. Roughly what this means is that if you make a small change - e.g. one edit in this example - then your cost function shouldn't change a lot.
If you don't have some form of continuity condition you are actually just doing random search.
