It's probably not that they can't classify fake accounts but that doing so at a scale requires too large resources and it's more economical to employ only simpl

It's probably not that they can't classify fake accounts but that doing so at a scale requires too large resources and it's more economical to employ only simpler techniques that weed out most of the bots and live with the remaining few. Then if they also happen to anyhow benefit from the existence of those fake accounts then it becomes even harder to justify costly filters and leads to a sufficient-enough-to-not-look-obviously-weak grade filtering