Those "engagement numbers" may be misleading
Though reading the crystal ball for these groups is a fools game, much like the low quality Russian campaigns from a few years ago, there are a couple of forks in the road:
One is they are busywork for unskilled operators whose leadership is still billing for their services. Which has happened.
Another is that the objectives of the campaigns aren't conventional engagement, and the producers know their transparent and shoddy work is not in itself convincing anybody.
Why persist at it then? Depending on the specifics of the campaign, it may be boosting the transmission of payload related to their posts but not linked to the accounts sending the low quality messages. This is one way to boost another campaign while avoiding simple network analysis. So if you want conspiracies about the Nord Stream pipeline or a dirty bomb, mentioning them in active but ignored campaigns may reach the often surprisingly low threshold to get boosted further by many of the "mystery box" algorithms that drive the modern online world.
Interestingly, the low quality posts can be a strategy, as they get filtered from the results but then can help lift a post from somebody in Gulfport or Billings that uses the same terms. So they may work as well as higher quality posts in biasing the results, but are made stealthier as they are either ignored when seen or filtered out. This is one of the ways people have amplified seemingly organic hate speech to help it get viral. It can assist posts reflecting it's talking points on other accounts not in it's engagement network, often without the intent or much visibility of the operator. One of the curses of using ML models as a magic wand for so many things is the opacity of their decision making process. And there is always a shit talking idiot out there, so why try to find and amplify them on your own when you can get Facebook or Twitter to do it for you, automatically?
Not sure how resistant either of those companies are today, but it was a brisk business gaming their naive weighting system just a few years ago. And now those systems aren't just there and Google and the Hyperscalers. Before the game was to springboard on one big platform to bounce the results on the other majors. Reddit to Twitter to Facebook or YouTube or vice versa depending on the crews. Now that so many other companies are using the same techniques, even as the majors start playing whack a mole and locking down direct actions on their platforms, the attackers just shifted to pulling the same stunt on systems that use the same mechanics but are not policed as effectively. Since the majors surface and scrape content from everywhere else, it still can go viral on their platforms even when the actors aren't running a campaign directly on their system.
Like matter creates gravity, these hyper-scale algorithmic systems create emergent instabilities and are prone to being hijacked. Systems like that are unstable and inherently dangerous. People didn't listen at first, because "how could Twitter be dangerous", then Facebook accelerated ethnic cleansing and genocide in south east Asia, because their black box was left to run unsupervised in a country where the developers weren't paying attention to anything other than the revenue and didn't speak the language. Villages burned and piles of corpses. Millions internally displaced, and many forced into permanent refugee camps in neighboring countries.
This stuff is not academic, it is neither funny nor fun.