Your objective is to select the line with the highest EV. Full stop.
If you have strong reasons to believe that your particular opponent will make specific types of mistakes under certain lines and board runouts, then you can think about making exploitative adjustments.
Most (approximate) GTO solutions play a large fraction of hands with a mixed strategy. So, for example, if you know that your opponent tends to overfold on certain rivers, this will tend to break the degeneracy in favour of particular actions -- your exploitative solution is to select only one action instead of playing the mixed strategy.
But in this case, the exploit is the new highest EV line.
So, in general, first determine your best guess as to the approximate GTO solution for hand and action to this point. Then, if and only if you have strong reason to believe in specific deviations from optimality of either the player pool in general, or this opponent in particular, do you work out the best exploitative adjustment to make.
I think most players overdo this last step -- they tend to overgeneralize from very small sample sizes and go out of their way to "exploit" opponents who may not be deviating from optimality as much as they might think. But that might just be the Bayesian in me talking ...
Aside: Confining your exploitative adjustments to only those hands played with a mixed strategy also helps to camouflage what you're doing -- your opponent will need a large sample to detect that you are playing certain lines more frequently than you "should".