[guided]The key geometric idea is to place the poles in the mirror so that the decomposition $S^n = A \cup B$ is compatible with the reflection. If we chose poles away from the mirror, $r$ would swap the two halves $A$ and $B$, and we could not use naturality to reduce dimensions.
With $N$ and $S$ in the hyperplane $\{x_1 = 0\}$, the reflection $r(x_1, x_2, \ldots, x_{n+1}) = (-x_1, x_2, \ldots, x_{n+1})$ fixes both poles, so $r(A) = A$ and $r(B) = B$. This means $r$ induces self-maps on $A$, $B$, and $A \cap B$, and the entire Mayer--Vietoris sequence is natural with respect to $r$.
The intersection $A \cap B = S^n \setminus \{N, S\}$ is the sphere with two antipodal points removed. The deformation retraction onto the equator $S^{n-1} = \{x \in S^n : x_{n+1} = 0\}$ is given by normalising the projection $(x_1, \ldots, x_n, x_{n+1}) \mapsto \frac{(x_1, \ldots, x_n, 0)}{|(x_1, \ldots, x_n)|}$. Under this retraction, $r$ restricts to $(x_1, x_2, \ldots, x_n) \mapsto (-x_1, x_2, \ldots, x_n)$ on $S^{n-1}$, which is again a reflection across a hyperplane through the origin.[/guided]