I think google can and does (and most certainly could) associate abstract concepts (word combinations) together. When you have the largest writing sample the world has ever known relevancy could be realized as a function of context and frequency. "flea brush" and "dog" probably appear together quite often (or frequently), whereas "flea brush" and "toothbrush" are probably not used together very regularly at all. Context I think is trickier - shoes is associated with socks and running, does this imply that socks and running are associated?
In any case when I link relevantly I look for sites most associated to my own and gradually branch outword. I think it more productive to define a lack of relativity then to decide what is relative enough.