A different approach is given by the following argument (not using Hilbert 90). Pick an element $\gamma\in K\setminus k$. Let $z$ be an arbitrary element of
$k$. Consider
$$
x=\frac{z+\gamma}{z+\gamma^q}\in K.
$$
We have
$$
\sigma(x)=\frac{\sigma(z)+\sigma(\gamma)}{\sigma(z)+\sigma(\gamma^q)}=\frac{z+\gamma^q}{z+\gamma}=\frac1x.
$$
Therefore $N(x)=1$. It is easy to show that different choices of $z$ yield different elements $x$, so there are at least $q$ elements in $\mathrm{ker} N_{K/k}$. Furthermore,
obviously $1\in \mathrm{ker} N_{K/k}$ and $x=x(z)\neq1$ for all $z\in k$. Therefore
$|\mathrm{ker} N_{K/k}|\ge q+1$. On the other hand
$$
N_{K/k}(x)=x^{q+1},
$$
so there cannot be more than $q+1$ solutions to the equation $N_{K/k}(x)=1$.
Note that the construction is a cousin of the parametrization of points on the complex unit circle (other than $z=1$) by the recipe
$$
z=\frac{x+i}{x-i},\quad x\in\mathbb{R}.
$$
Admittedly Hilbert 90 also has a lot of appeal here.