We present a theoretical framework for understanding the relationship between anchoring bias, hypothetical bias, and cheap talk in contingent valuation surveys. In our theory, interviewers provide agents with signals such as cheap talk and bid values while eliciting the value for nonmarket goods. In response to these signals, agents revise their prior distributions over the value of the good. Previous empirical studies have failed to account for the interaction between cheap talk and anchoring during this updating process, leading researchers to incorrectly assess the effects of cheap talk in reducing hypothetical bias. In particular, we predict that cheap talk will appear to be more effective for relatively large bids. We test our theory in an experimental setting where agents are asked to make a hypothetical, voluntary contribution to a public good. The experimental results, as well as several recent empirical studies, are consistent with the theory.