I think it'll end up like Facebook (the social media platform, not the company). Eventually you'll hit model collapse for new models trained off uncurated internet data once a critical portion of all online posts are made by AI, and it'll become Much more expensive to create quality, up-to-date datasets for new models. Older/less tech literate people will stay on the big, AI-dominated platforms getting their brains melted by increasingly compelling, individually-tailored AI propaganda and everyone else will move to newer, less enshittified platforms until the cycle repeats.
Maybe we'll see an increase in discord/matrix style chatroom type social media, since it's easier to curate those and be relatively confident everyone in a particular server is human. I also think most current fediverse platforms are also marginally more resistant to AI bots because individual servers can have an application process that verifies your humanity, and then defederate from instances that don't do that.
Basically anything that can segment the Unceasing Firehose of traffic on the big social media platforms into smaller chunks that can be more effectively moderated, ideally by volunteers because a large tech company would probably just automate moderation and then you're back at square 1.