This study analyzes the factors for detecting agents to implement a human-like agent in werewolf games, which are new themes in game studies of artificial intelligence. A comparative experiment was conducted to reveal the influences regarding the degree of disclosure of the agent existence; seven players and an agent participated in four games. Communication logs were collected from the game logs. Impressions for each player was reported in questionnaires with five-point scales and detection of the agent were reported in a free writing. Although the most players, who did not know that an agent played the games, did not be aware of the true character of the agent, the most players who knew the existence could detect the agent. The statements of detected agents differed from them of other players. This study concludes that the technical requirement for agents is to adapt to statements of other players.