Forecast guidance on weather and floods derived from expert system is increasingly deployed by environmental agencies and commercial companies. As artificial intelligence (AI) emerges as a useful tool in providing recommendations in this situation, trust plays a critical role in building effective interactions between human decision-maker and AI. Chatbot, an artificial intelligence robot conducting conversation with human, becomes increasingly popular in contemporary websites and applications in delivering recommendations made by algorithms. There is little evidence about how this conversational robot affects human trust towards the recommendations. In this study, we investigate the role of chatbot in the human-machine interaction in a cost-loss game using lab experiments. The chatbot’s main role is to give advice to the human decision maker whether to take the risk of a possible loss or pay a cost to avoid the risk. We exogenously manipulate the chatbot’s ability in conducting the conversation in a novel experimental system, and find that participants’ subsequent decision to follow the chatbot’s advice is systematically influenced by its chat ability. In addition, we find that the influence of chat ability on human decision is mainly mediated by the perceived intelligence of the robot, rather than emotions or social distance. The policy implications of the study in environmental risk communication will also be discussed.