1. Don't call up a plot of the key words based on more than one text file. It doesn't make sense! Anyway the plot will only show the words in the first text file. If you want to see a plot of a certain word or phrase in various different files, use Concord dispersion.
2. There can be no guarantee that the "key" words are "key" in the sense which you may attach to "key". An "important" word might occur once only in a text. They are merely the words which are outstandingly frequent or infrequent in comparison with the reference corpus.
3. Compare apples with pears, or, better still, Coxes with Granny Smiths. So choose your reference corpus in some principled way. The computer is not intelligent and will try to do whatever comparisons you ask it to, so it's up to you to use human intelligence and avoid comparing apples with phone boxes!
If it didn't work...
For the procedure to work, a number of conditions must be right: the language defined for each word list must be the same (that is, Mexican Spanish and Iberian Spanish count as the same but Iberian Spanish and Brazilian Portuguese count as different so could not be compared in this process); each word list must have been sorted alphabetically in ascending order before the comparison is made. (The program tries to ensure this, automatically.) Also, any prefixes or suffixes must match.