Optimal targeting of social aid is a key issue in the design of public policy. A key aspect of policy design is to create welfare systems that are manipulation-proof. We study demand-side manipulation of a large, nationwide welfare program in Georgia. We use a rich tapestry of administrative and survey data to study a group of welfare-recipient households, whose total baseline income is roughly 100 USD per month. We start by documenting sizable jumps in household manipulation behavior at a key benefit discontinuity. We build a Becker (1968) style model of manipulation, which we use to inform our empirical strategy – a fuzzy regression discontinuity-instrumental variables design. Using administrative data, we find that those in households that engage in manipulation subsequently also engage more in the formal labor market, particularly men. We explore various channels for this finding. Using a marginal treatment effect analysis, we rule out that households are selecting on the unobserved (labor market) gains to manipulation. Manipulating households consequently have higher levels of income, from welfare and the labor market. Using survey data that we collect, we find that welfare-manipulating households spend the vast majority of their additional income on children, increasing expenditure on clothes and education. Given that adults in manipulating households are working more and have more income it is not clear a priori whether childhood outcomes will improve or not, given the countervailing inputs of parental time and money. With an early childhood skill production model in mind, we study a battery of outcomes for children and young people in the household. We find no significant changes for health outcomes for 0-5 year olds, nor do we find changes in educational engagement for teenagers. We do find an increase in university attendance for young people still living at home, particularly young women.