Incentivising agent performance is a double-edged sword: while it can encourage agents to perform better, it might also nudge them into cheating and manipulating results to their benefit. In this VoxDevTalk, Guojun He discusses his work with Michael Greenstone, Ruixue Jia, and Tong Liu on this classic principal-agent problem in the context of how Chinese local governments self-report meeting air pollution-reduction targets imposed (and incentivised) by the central government. An analysis of these reports reveals evidence of significant under-reporting by local governments before the central government installed automated real-time pollution monitoring devices across the country. Under-reporting was larger in areas with higher levels of actual pollution, ostensibly since these local governments face greater challenges in meeting pollution-reduction targets. How accurately local governments report pollution figures also has impacts on public welfare, with people exposed to pollution information more likely to search for information on face masks and air filters. Biased information thus prevented people from optimally protecting themselves prior to the introduction of automation.