Of course, it is false; yet, it is true in a certain sense!
$\det(B-\lambda I_{2n})=\det(\lambda^2I_n-AA^T)$; then $\rho(B)^2=\rho(AA^T)$ and it suffices to consider $\rho(AA^T)$.
i) Consider $E=I_{10}+{J_{10}}^T$ where $J_k$ is the nilpotent Jordan block of dimension $k$. Then $E$ contains $9$ non-zero entries under the diagonal and $\rho(EE^T)\approx 3.91$. Consider $C=I_{10}+R$ where $r_{i,j}=0$ except $r_{8,1}=r_{9,1}=r_{9,2}=r_{10,1}=r_{10,2}=r_{10,3}=1$. Then $C$ contains $6$ non-zero entries under the diagonal and $\rho(CC^T)\approx 6.90$. That implies that your conjecture is false.
ii) Note that $A$ and $AA^T$ are non-negative matrices (in the sense $A\geq 0$ iff for every $i,j$, $a_{i,j}\geq 0$). Now, we transform one $0$ under the diagonal of $A$ in $1$; if $A'$ is the obtained matrix (it has a zero less than $A$), then $A'A'^T-AA^T\geq 0$ and, consequently (it is a theorem about the non-negative matrices!), $\rho(A'A'^T)\geq \rho(AA^T)$.
EDIT. What is important is the location of the $1$'s. If we replace some $0$'s by $1$'s far from (resp. near) the diagonal of $A$, then $\rho(AA^T)$ increases quickly (resp. slowly).