There is a difference between background-level bulk sniffing and someone-here-maybe-incited-violence targeted sniffing. The former is data collection, which is passive in the form practiced by “the feds”. The latter is data connection, putting effort into connecting a subset of the data that has been collected to form a story. Data connections need a framing, a nucleation seed, an impetus for why the feds might think such a connection is interesting or relevant or worth adding to their story about a larger incident. Collecting data is cheap and done in bulk, partly because it can be done passively and partly because the US govt paid a lot of money on storage and collection mechanisms. Connecting data is something that requires a lot more time, effort, patience, and vetting to make sure you are doing it right.
Or you can give the job to generative AI and hope it doesn’t hallucinate that someone innocent is guilty; with a large enough data pool (ie the internet, reality, what-have-you) it’s possible to select a misleading subset to support whatever hallucination you want.
It’s easy to do wrong, which is exactly why you don’t want the feds sniffing around. Especially now that they have the tools to automate doing it wrong, and might not know how to use them yet.