Suppose $A$ is a random subset of a finite group $G$, such that each element of $A$ independently belongs to $G$ with probability $p$ (in your case $G = S_n$). One can see that $\forall H \leq G (P(A \subset H) = (1 - p)^{|G| - |H|}$. It is also true, that $P(A \subset H) = \Sigma_{K \leq H} P(\langle A \rangle = K)$. Thus, $P(\langle A \rangle = H) = \Sigma_{K \leq H} \mu(H, K)P(A \subset K) = \Sigma_{K \leq H} \mu(H, K)(1 - p)^{|G| - |K|}$, where $\mu$ is the Moebius function for subgroup lattice of $G$. Thus we have the formula:
$$E|\langle A \rangle| = \Sigma_{H \leq G} \Sigma_{K \leq H} |H|\mu(H, K)(1 - p)^{|G| - |K|}$$
That is one of the possible forms of answer, despite it is quite hard to anyhow simplify it, as one must know the structure of subgroup lattice of $G$ to calculate the Moebius function (In your case it is especially hard as the subgroup lattice of $S_n$ is not well described).