[guided]Let $\beta+1 \leq \alpha$, and suppose the previous levels agree:
\begin{align*}
L_\beta^M=L_\beta.
\end{align*}
At a successor stage, the constructible hierarchy does not take a union; it forms the subsets of the previous level that are first-order definable over that previous level. Thus $M$ computes
\begin{align*}
L_{\beta+1}^M=\operatorname{Def}^M(L_\beta^M,\in),
\end{align*}
while the ambient universe computes
\begin{align*}
L_{\beta+1}=\operatorname{Def}(L_\beta,\in).
\end{align*}
By the induction hypothesis, the underlying set of the structure is the same in both computations. So the problem reduces to showing that $M$ and the ambient universe agree about which subsets of $L_\beta$ are definable over $(L_\beta,\in)$.
Fix a subset $A \subseteq L_\beta$. Externally, $A$ is in $\operatorname{Def}(L_\beta,\in)$ exactly when there are a first-order formula $\varphi(v,w_1,\dots,w_n)$ in the language $\{\in\}$ and parameters $a_1,\dots,a_n \in L_\beta$ such that
\begin{align*}
A=\{x \in L_\beta : (L_\beta,\in)\models \varphi(x,a_1,\dots,a_n)\}.
\end{align*}
We now check that all data in this definition are also available inside $M$. The formula $\varphi$ is represented by a natural-number code, and the relevant fragment of ZF in $M$ includes the coding of syntax; since $M$ is transitive and satisfies the needed Infinity fragment, these finite syntactic codes are the same in $M$ as externally. The parameters $a_1,\dots,a_n$ lie in $L_\beta$, and $L_\beta=L_\beta^M$ by the induction hypothesis, so they are parameters available to $M$. Finally, the structure $(L_\beta,\in)$ is the same structure in both contexts: transitivity of $M$ makes membership between elements of $L_\beta$ absolute.
The remaining issue is truth in this set-sized structure. By the assumed satisfaction-coding fragment, $M$ has the satisfaction relation for $(L_\beta,\in)$, and this coded satisfaction relation agrees with the ambient one. Therefore, for every $x \in L_\beta$,
\begin{align*}
M \models ``(L_\beta,\in)\models \varphi(x,a_1,\dots,a_n)''
\end{align*}
if and only if
\begin{align*}
(L_\beta,\in)\models \varphi(x,a_1,\dots,a_n).
\end{align*}
Thus the same formula and the same parameters select exactly the same elements $x \in L_\beta$ inside $M$ and externally. Since $M$ has the required Separation fragment for formulas defining subsets of earlier levels, $M$ forms precisely that subset at the successor stage. This proves
\begin{align*}
\operatorname{Def}^M(L_\beta,\in)=\operatorname{Def}(L_\beta,\in),
\end{align*}
and hence
\begin{align*}
L_{\beta+1}^M=L_{\beta+1}.
\end{align*}[/guided]