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