The issue wasn't the analytics either. The issue was the engagement algorithms and lack of accountability. Those problems still exist today.
We need to figure out how to target the malicious individuals and groups instead of getting creeped out by them to the point of destroying most of the so praised democratizing of computing. Between this and locking down the local desktop and mobile software and hardware, we've never got to having the promised "bicycle for the mind".
If people had perfect self-control, they wouldn't do it. IMO it's somewhat irresponsible for the algorithm makers to profit from that - it's basically selling an unregulated, heavily optimized drug. They downrank scammy content for instance, which limits its reach - why not also downrank trolling? (obviously bc the former directly impacts profits, but not the latter, but still)
You can legitimately claim that people respond in a very striking and predictable way to being set on fire, and even find ways to exploit this behavior for your benefit somehow, and it still doesn't make setting people on fire a net benefit or a service to them in any way.
Just because you can condition an intelligent organism in a certain way doesn't make that become a desirable outcome. Maybe you're identifying a doomsday switch, an exploit in the code that resists patching and bricks the machine. If you successfully do that, it's very much on you whether you make the logical leap to 'therefore we must apply this as hard as possible!'