Because the $t$-distribution is always narrower.

Because the $t$-distribution accounts for the additional uncertainty when using the sample standard deviation $s$ instead of $\sigma$.

Because the $t$-distribution is only used for very large datasets.

Because the $z$-distribution is only for proportions.