Let $\sigma \in A_n$ such that the disjoint cycle decomposition of $\sigma$ contains no even cycles or cycles of the same length. Although there is one conjugacy class of $\sigma$ in $S_n$, there are two conjugacy classes of $\sigma$ in $A_n$. We can choose any permutation in the original $S_n$ conjugacy class to represent one of the two $A_n$ classes. But how can we find another permutation to represent the other $A_n$ class? These answers suggest the following to be true:
Let $\sigma \in A_n$. If the conjugacy class of $\sigma$ splits in $A_n$, then $\tau \sigma \tau^{-1}$ is not conjugate (in $A_n$) to $\sigma$ for all $\tau \in S_n \setminus A_n$.
What is a proof of this fact?
From here we know that $\tau \sigma \tau^{-1} \neq \sigma$ for all odd $\tau$. But proving $\tau \sigma \tau^{-1} $ is not conjugate (in $A_n$) to $\sigma$ seems to be a stronger claim.