While poisoning an ARP cache, it is the legitimate user whose reply is received first rather than the attacker's fake reply when a corresponding request is sent. Maybe because the attacker is dependent on the user level program to sniff the request and generate the reply.