Random captchas ("We need to re-confirm you're a human") sounds like a great idea, although annoying to someone just wanting to post a quick 20-word tweet.

But it would have the useful side effect of slowing people down from tweeting impulsively, a practice that demeans the conversation and sometimes ends careers.

this would not work well because the sample would be biased

newer accounts are much more likely to be bots compared to older ones

you would have to sample IDs by age and create a distribution based on ID age

And then you can extrapolate how many overall are fake

This may work