Say we have $$\Box A = J$$
and $$\nabla \cdot A = 0\;.$$
Then $$0 = \Box (\nabla \cdot A) = \nabla \cdot J\;.$$
But, $$\nabla \cdot J - \partial_t \rho = 0\;.$$
So $$ \partial_t \rho = 0\;.$$
Thus, $$\rho = const.$$
But that doesn't make sense to me, especially when you consider a nonlinear system like Dirac-Maxwell. Where is my thinking wrong?
EDIT: As the answers point out the confusion stemmed from the fact that a gauge was already fixed when writing $\Box A = J$. However I do want to point out that orginially I was thinking about Coulomb gauge in which one has
$$\Box A = -\mathbb{P} J$$
where $\mathbb{P} = I - \nabla \Delta^{-1} \nabla$ projects onto divergence free vector fields since in the Coulomb gauge
$$\Box A = -J + \nabla \partial_t V$$
but $\nabla \partial_t V$ is the rotation-free part of $J$ as a quick calculation shows. In this gauge, taking the divergence of $\Box A = -\mathbb{P} J$ is not problematic since one applies $\nabla \cdot$ to the divergence free part of the current only. The rotation-free part of $J$ has nonzero divergence.