[guided]We want to arrange the sets so that $A \cap B$ is non-empty but strictly smaller than $A$. Why? Because we are performing induction on $|A|$, and the inductive step will apply the hypothesis to the pair $(A \cap B, A \cup B)$, which requires $|A \cap B| < |A|$.
First, translate $A$ so that $0 \in A$ (pick any element of $A$ and subtract it). Now we need to translate $B$ so that $0 \in B$ (ensuring $A \cap B \neq \varnothing$) while also ensuring some element of $A$ stays outside $B$ (ensuring $A \cap B \subsetneq A$).
Pick any $a \in A$ with $a \neq 0$ (possible since $|A| \geq 2$). Since $p$ is prime and $a \neq 0$, the element $a$ is a generator of the cyclic group $\mathbb{Z}_p$: the multiples $0, a, 2a, \ldots, (p-1)a$ exhaust all of $\mathbb{Z}_p$. In particular, as $k$ ranges over $\{0, 1, \ldots, p-1\}$, the translates $B - ka$ cycle through all shifts of $B$ by multiples of $a$.
Consider the set $B$ and the arithmetic progression $0, a, 2a, \ldots$. Since this progression hits every element of $\mathbb{Z}_p$ and $B \subsetneq \mathbb{Z}_p$ (we have $|B| \leq p - 1$ because $|A| \geq 2$ and $|A| + |B| \leq p + 1$), there exists a smallest $k \geq 0$ such that $ka \in B$ but $(k+1)a \notin B$. Translating $B$ by $-ka$, we ensure $0 \in B$ and $a \notin B$.
After these translations: $0 \in A \cap B$ (so $A \cap B \neq \varnothing$), and $a \in A \setminus B$ (so $A \cap B \subsetneq A$). This is exactly the configuration needed for the inductive step.
Note where primality was used: the claim that $a$ generates $\mathbb{Z}_p$ requires $\gcd(a, p) = 1$, which is automatic when $p$ is prime and $a \not\equiv 0$. In a composite group $\mathbb{Z}_n$, a non-zero element $a$ with $\gcd(a, n) > 1$ generates a proper subgroup, and the argument would fail.[/guided]