Cheers to everyone. I' ve got a serious doubt about the following: consider the annihilation operator $\hat a$. For practical reasons, I sometimes find useful redefining it in the following way : $\hat a' =\hat a e^{i \phi}$, with $\phi \in \mathbb R$. If I add a new global phase to each eigenstate of $\hat a^\dagger \hat a$, $| 1 \rangle \rightarrow | 1 \rangle e^{i \phi}, \quad | 2 \rangle \rightarrow | 2 \rangle e^{2 i \phi} \,\dots$, I have a new annihilation operator $\hat a'$ and a new equivalent Hilbert space.
Is this $\hat a'$ physically reliable? Consider the time evolution of a state with Hamiltonian $\mathcal H = \alpha \hat a + \alpha^* \hat a^\dagger$, with $\alpha \in \mathbb C$. With the transformation described above $\alpha$ can be considered to be real without loss of generality. Is this correct?