While the goals are slightly different, I highly recommend reading[1] about the methods used by GCHQ's JTRIG unit (including the full Snowden document[2]). They are not really about
specific metho9ds to manipulate groups of people, but the general ideas behind solving that type of problem.
A particularly important topic discussed in [2] is "Attention Management". The first step humans use to process sensory input is deciding which (small) part of that data is important. We end up blatantly ignoring[3] most sensory data, because our capacity for attention is surprisingly small and very easily manipulated[4] if you know how to flood someone's input buffers[5].
The more troubling problems start when it isn't a human deciding how to manipulate attention, but instead it's a bunch of machine learning trying to "optimize engagement" (attention).
[1] https://theintercept.com/2014/02/24/jtrig-manipulation/
[2] https://theintercept.com/document/2014/02/24/art-deception-t...
[3] how long does it take you to notice the difference between these two flickering frames: https://www.cse.iitk.ac.in/users/se367/10/presentation_local...
[4] https://www.youtube.com/watch?v=GZGY0wPAnus
[5] Apollo Robins is shockingly good at using banter to DoS someone's attention: https://www.youtube.com/watch?v=dTa7rC1oUnk , https://youtu.be/1kkOKvPrdZ4?t=1882