The $t$-distribution is always narrower.

The $t$-distribution accounts for the uncertainty of using the sample standard deviation $s$ as an estimate of $\sigma$.

The $t$-distribution is only for proportions.

The $t$-distribution does not require a random sample.