Our algorithm optimizes for CTR context, meaning – different items can have high CTR in different contexts than others. Contexts can be different sites, different content users are consuming, different user segments, etc. So, specific items might have very high CTR in a very specific contexts but low CTR elsewhere, and will not get a lot of impressions. Have we given them more impressions – their CTR would drop as they would show up in contexts where their CTR is lower.
The important thing to note here is that context matters. There's no single item that performs the best in all contexts. This is actually one reason why our algorithm works well, because it takes the context into consideration and does not simply show the highest averagely performing item everywhere.