This scrapes full stories from AO3 for natural language processing purposes!
As far as I can tell, it doesn't break their terms of service.
It's built on https://github.com/ArmindoFlores/ao3_api , which does a good job of grabbing individual works' metadata but doesn't get the actual texts.
It should turn a work_dictionary into a folder of text files with a nice, https://github.com/dhmit/gender_analysis - ready csv file.
Warning: extremely hacked together, untested.