We consider a standard model of cheap talk when the receiver has the option to verify the sender’s message at some cost. Verification entails a trade-off for the receiver: learning the state perfectly versus not paying the cost of verifying. An equilibrium partition consists of a mix of verified and unverified intervals, similar in structure to Bijkerk et al. (2018). We analyze a subset of equilibria with exactly one unverified interval. Fixing the informativeness of the equilibrium and the bias between sender and receiver, we show that, for a small bias, joint welfare is non-monotonic in the cost of verification, and for larger biases, it is always increasing in the cost of verification. Further, when costs decrease to the point where informativeness increases, welfare increases discontinuously. Finally, joint welfare is also non-monotonic in the bias between the sender and receiver. Our results suggest that the introduction of a verification technology may have both non-monotonic and discontinuous effects on societal welfare, depending on its cost and the conflict of interest between the agents.
