This study examines how South Korean Twitch streamers construct narratives through real-time conversations with ChatGPT. I draw on a corpus of ten livestreams featuring Korean streamers broadcasting their conversations with ChatGPT, in which they read out the messages that they exchange with ChatGPT in real time and provide commentary on the ongoing interaction for their viewers. Focusing on three excerpts, each taken from a different video, I analyze the livestreams from a narrative perspective to demonstrate how streamers use their continuous evaluation of ChatGPT’s dialogue to construct “small stories” (Bamberg and Georgakopoulou 2008; Georgakopoulou 2006). I show how various linguistic and paralinguistic resources accomplish this evaluation, such as tone, pace of speech, and evaluative indexicals. Drawing on Bamberg's (1997) theory of narrative positioning, I also demonstrate that these small stories often position the streamer vis-a-vis ChatGPT in antagonistic relationships, by showing streamers either being critical of the dialogue ChatGPT produces or making it out to be a demanding or difficult interlocutor. My analysis finds that many of these small stories of conflict are connected to “master narratives” (Tannen 2008) of “AI attacking humans,” or "AI dominating humanity" by positioning ChatGPT as an aggressor vis-a-vis the streamer as the victim; the invocation of such master narratives allows streamers to further heighten the tellability of their stories and engage their viewers in their livestreams by connecting to representations of this master narrative in popular works of fiction. This study contributes to our understanding of the interactional work speakers do to imbue their interactions with chatbots with meaning, and of the ways in which master narratives of “AI domination of humanity” are brought into everyday talk about AI chatbot conversations to apply to real, not fictionalized, interactions with technology.