Some fields of economics, studying health and well-being for instance, have so far relied on non-incentivized survey data, mostly because the survey answers are unverifiable. It has long been thought that it is impossible to incentivize people to truthfully report something as subjective as how they feel, or how happy they are. Over the past fifteen years, several methods have been proposed to incentivize truth-telling for such unverifiable questions, the best-known being the Bayesian truth-serum of Prelec (2004). We test the impact of truth-telling incentives in a large-scale, online experiment about health and well-being. We exogenously introduce a quantifiable bias in responses using defaults. We find that, in the presence of a default, incentives induce more effort, which reduces the default bias. We also find that, in the absence of default, data quality is good and incentives have no impact.