Scale is large but these guys have database level access.
In the end, all Twitter, Facebook, etc. are is a huge database.
If you can imagine a database as large as Twitter, it's a small jump to assume you can grep it.
They can just watch for certain topics they want to see starting to trend.
Once they see it, they send it up the chain of command to get direction on what to do.
Note how we can get things trending, hashtags and such, and it takes time for them to be removed.
Sometimes things aren't removed, they are redirected.
Said another way:
If it were me, I'd design a defensive surveillance layer that watches for a long list of keywords.
Upon keyword spotted, I'd create a strategy layer that would review to identify if there is opportunity to redirect, plant disinfo, or even to promote a path that I already know is off course.
If a topic can not be redirected, then I'd flood it with shills or AI shills to disrupt.
On top sensitive topics it would go up another layer to management for direction, along with recommendations for what could be done.
In some cases, and on topics that hit too close to home (think wikileaks-level releases) I'd revert to full scale DDOS or takedowns.
This is pretty much the nuclear option, so while there would be standing orders to flood some topics a full net takedown would be rare and heavily considered/coordinated with friends and allies.
On offense, I'd have a program for outbound promo of topics I want trending.
I'd have an allied thought leader (e.g., NYT) introduce a topic.
I'd jump on it and flood the trend.
I'd use the combined expertise of several universities-worth of psychologists to hone my tactics on psych momentum.
I can create stars this way.
I can create trends.
I can create whole social movements.
Pic related to offensive capabilities.
Example: Who thought up the pussy hat? How did he make it popular? This was a power demonstration, a big attempt to warn POTUS not to mess with the swamp.