Online reviews are typically written by volunteers and, consequently, accurate information about seller quality may be underprovided. We study the extent of this under-provision using a randomized experiment conducted by Airbnb. In the treatment, buyers are offered a coupon to review listings that have no prior reviews. The treatment induces additional reviews, which are more negative on average. Induced reviews do not change nights sold, although they affect the types of transactions that occur. Measures of transaction quality for treated sellers do not improve. We show how market conditions and the design of the reputation system can explain our findings.