For $k < 2$ and $k > \dim V - 2$ it's always true that a $k$-form $\omega$ both is decomposable and satisfies $\omega \wedge \omega = 0$.

For $k = 2$ (and provided the field underlying $V$ does not have characteristic $2$), the condition $\omega \wedge \omega = 0$ is both necessary and sufficient for decomposability. (Proving this is a nice exercise.)

The converse is not true in general, however. If $k$ is odd, then all $k$-forms $\omega$ satisfy $\omega \wedge \omega = 0$, but not all odd-degree multivectors (or, dually, forms) are decomposable:

**Example** If $\dim V \geq 5$, pick a basis $(E_a)$ and denote the dual basis by $(e_a)$. Then, the $3$-form
$$\psi := (e^1 \wedge e^2 + e^3 \wedge e^4) \wedge e^5$$
satisfies $\psi \wedge \psi = 0$ but we can show that it is indecomposable: Contracting a vector into a decomposable form yields a decomposable form. On the other hand, $$\iota_{E^5} \psi = e^1 \wedge e^2 + e^3 \wedge e^4 ,$$ and computing gives $(\iota_{E^5} \psi) \wedge (\iota_{E^5} \psi) \neq 0$, so by the criterion for $k = 2$, $\iota_{E^5} \psi$ is indecomposable.

For an algorithm that checks decomposability of a general $k$-form, see this old question.

**Remark** For $2 \leq k \leq \dim V - 2$, most $k$-forms are not decomposable, and we can quantify this assertion: For $0 \leq k \leq \dim V$, any $k$-vector $E_{a_1} \wedge \cdots \wedge E_{a_k}$ determines a $k$-plane in $V$, namely, $\langle E_{a_1}, \cdots, E_{a_k} \rangle$, and any $k$-plane in $V$ determines an underlying form up to an overall nonzero multiplicative constant. So, we may regard the space $D_k(V)$ of (nonzero) decomposable $k$-forms as a (punctured) line bundle over the space of all $k$-planes in $V$; this latter space is called the *Grassmannian (manifold)*, $Gr(k, V)$, and it has dimension $k (\dim V - k)$, so $D_k(V)$ is a smooth manifold of dimension $k (\dim V - k) + 1$. On the other hand, the space of all $k$-forms has dimension $n \choose k$, and for $2 \leq k \leq \dim V - 2$,
$$\dim D_k(V) = k (\dim V - k) + 1 < {\dim V \choose k}$$ (but note that equality holds for $k = 1, \dim V - 1$).

Alternatively, the Plücker embedding realizes $Gr(k, V)$ as a projective variety in $\Bbb P(\Lambda^k V)$, and when $2 \leq k \leq \dim V - 2$ it is a proper subvariety, so its complement is nonempty and Zariski-open (and hence, when the underlying field is $\Bbb R$ or $\Bbb C$, dense with respect to the usual topology).