I study the optimal algorithmic disclosure in a lending market where lenders use a predictive algorithm to mitigate adverse selection. The predictive algorithm is unobservable to borrowers and uses a manipulable borrower feature as input. A regulator maximizes market efficiency by disclosing information about the statistical properties of variables embedded in the predictive algorithm to borrowers. Under the optimal disclosure policy, the posterior belief consists of two disjoint regions in which the borrower feature is more relevant and less relevant in predicting borrower quality, respectively. The optimal disclosure policy differentiates posterior lending market equilibria by the equilibrium data manipulation levels. Equilibria with more data manipulation hurt market efficiency, but also discourage lenders’ use of the borrower feature. Equilibria with less data manipulation benefit from that and generate more efficient market outcomes. Unconditionally, the borrower feature is used less intensively under optimal disclosure. This information design problem can be reduced to a one-dimensional maximization problem by imposing a mild distributional assumption on manipulation cost. As an extension, I also discuss the joint design of algorithmic disclosure and costly verification.