AI-driven chatbots are increasingly used in language learning to offer personalized practice and promote learner autonomy. However, chatbot interactions are typically individual and relatively new, leaving gaps in understanding how learners engage with these tools. This study analyzes interactions between engineering students and a Poe-based chatbot designed to act as an investor evaluating student proposals. The goal is to help students prepare for live assessed simulations. The study employs the concept of 'field' from systemic functional linguistics, or 'what the conversation is about.' It specifically examines 'field shifts,' when learners deviate from the main topic, affecting conversation quality. Field shifts can lead to incomplete communication or unmet objectives. Through qualitative analysis, the study identifies several types of field shifts, with 'Zooming Out' (providing overly general answers), 'Wide Miss' (responses that broadly miss the main topic), and 'Blurred Shot' (fragmented, unclear responses) being the most common. These types of shifts often reduced coherence and negatively impacted learners' ability to achieve their communicative goals. This session targets language instructors and instructional designers interested in chatbot-mediated learning. Understanding common field shifts can help them design more effective chatbot prompts and scaffolding strategies, ultimately fostering more focused and goal-oriented learner interactions.