Significant difference between response to real and hypothetical valuation questions is often referred to as hypothetical bias. Some economists have had success with using "cheap talk" (which entails reading a script that explicitly highlights the hypothetical bias problem before participants make any decisions) as a means of generating unbiased responses in a referendum format. In this article, we test the robustness of cheap talk using a voluntary contribution mechanism with a provision point over a wide range of possible payment amounts. Our results confirm the existence of hypothetical bias, and suggest that cheap talk may eliminate hypothetical bias, but only for respondents facing higher payments.