Start with classic conditioning that’s implemented in every large social media platform and go from there.
Or go in reverse, look at research into correlation between mental health issues and social media use and extrapolate contributing factors, from those extract the features
Should give a starting point for nailing down the definition.
See: “it’s just an init system”where it’s now also a resolver, log system, etc.
I can buy good intentions, but this opens up too much possibility for not-so-good-intended consequences. Deliberate or emergent.