Comparing Percentages: Z-Score for Population Proportions or Chi-square?

#1
Hello,

I am doing a linguistic analysis using LIWC (Linguistic Inquiry Word Count), which is a software that for each text you input, gives you an output composed of 95 grammatical categories and tells you which percentage of the overall text belongs to that category. I am using it to compare the speech of Innocent and Guilty people coming from an in-lab experiment.
An example of a part of the output:

Guilty: Personal Pronouns (24%)
Innocent: Personal Pronouns (22%)

I also have all the counts, but the tricky thing is that since the innocent people talked a lot less than the guilty ones, I have two different n-sizes. In another study I had used logistic regression, but in that occasion both n had been very similar. Researching online I have found as possible solution the Z-score for Population Proportions, although other people mention a Chi-square (although I'm not sure what the expected population would be in this case).

Any insight on the best way to approach this would be really appreciated :)