In full-length RNA-seq, given two genes have same expression level, then there is indeed a correlation between counts and length, as longer genes produce more fragments. This of course assumes that all fragments can be sequenced the same, like PCR/GC/mappability-bias was not existing which is of course not true, so this statement is an approximation, not a precise reality, like everything in biology. Maybe this sentence above is a typo, so if a gene is 20% shorter it should roughly produce 20% fewer reads, given expression level was similar.