Bots are autonomous software agents able to imitate human behaviour which makes them interesting for interactive processes such as electronic negotiations. In electronic negotiation training, humans often negotiate with negotiation software agents which respond quickly to the offers of the human participants. Currently, these agents are limited in their communication behaviour and thus restrain the effectiveness of electronic negotiation training. For an effective training, coherent and transparent communication processes are desirable, in which the agent takes up the human’s arguments and provides their own reasonable arguments. Following the design science research methodology, we derive requirements and a meta-design for a negotiation bot to improve communication quality, and finally present our newly developed negotiation bot. The evaluation comparing the bot with an existing agent shows that although the bot sometimes provides unsuitable arguments, the bot imitates human behaviour well and ensures coherent communication processes. The bot can thus improve communication training for electronic negotiations.