If most accounts aren't people, then what most people do on the service is likely to have nothing in common with what most accounts do on the service.
There is also the orphan account situation. My twitter and facebook and linkedin accounts are unused but not deleted. I would theorize that's yet another problem, once you cross off the list of bots, if the majority of remaining accounts are unused, that means what the majority of accounts do has little to no relationship to what actual active users do. For marketing and PR purposes I'm sure my unused accounts are being monetized and proudly being counted as a "user", which is a whole nother problem.
There is an even deeper reason the analysis is wrong. Inspired by writing this post I logged into facebook for the first time in months. My kids stopped using it years ago and now posting is very power law distribution where virtually all my timeline posts are now from exactly two people who post or repost something roughly every waking hour and nobody else can keep up with their meme-stream which floods everything else out. When a sample is power law skewed like facebook, words like average and median are irrelevant and meaningless. Power law distributions have other ways to analyze them like half life / doubling intervals and talking about the actual exponent of the distribution and stuff like that. The average post "on facebook" comes from precisely two meme-crazy shitposters I feel familial / workplace obligation or peer pressure to friend, but the average user's post is an entirely different phenomena and is a perhaps annual event at most. Oh look its been a year since my last summer vacation, here's my annual summer vacation post.
There is a similar problem with email, yes 99.9% of email sent is spam no one reads. No that does not mean no one uses email or there are no important emails.