Crowdfunding platforms allow consumers and firms interact even before new product launch, and enable entrepreneurs and investors to interact with fewer geographic constraints. Such crowdfunding platforms provide entrepreneurs opportunities to raise funds for their creative ideas. However, the information asymmetry could distort the market participation and leads to sub-optimal outcomes.  To mitigate such distortion, crowdfunding platforms often allow fundraisers to provide various (but unverified) information about the focal projects. On one hand, these unverified descriptions of initiators’ characteristics and project prospects (besides the verified information such as credit rating) may help funders evaluate the underlying projects, while on the other hand could result in cheap talk by the entrepreneurs that reduces the credibility of the information about the underlying projects. We construct a structural model of cheap talk and utilize detailed online crowdfunding transaction data coupled with supervised machine learning for the project descriptions, to study the informational impact in this market. We show that cheap talk, on average, does not improve the crowdfunding outcome, though increases the funding prospects. However, the effect of cheap talk on crowdfunding is heterogeneous over projects. While “cheap talk” helps low credit rating initiators in crowdfunding, this positive effect depreciates as the initiators’ credit ratings improve. Interestingly, counterfactual experiments show “paid cheap-talk” could actually benefit everyone. We then discuss the related managerial implications for platforms and entrepreneurs.