This paper proposes a framework for visual target navigation in unmanned aerial vehicles (UAVs) that leverages LLMs. By utilizing the powerful language understanding capabilities and prior knowledge of LLMs, this framework enhances UAVs’ comprehension of natural language instructions and their ability to plan tasks. Initially, the method integrates task instructions, contextual backgrounds, and the UAV’s functional library through designed prompt information. Subsequently, LLM serves as a planner, generating long-term plans that align with the task instructions and selecting corresponding functional modules based on the UAV’s capabilities to execute control. Leveraging the strong reasoning abilities of LLMs, we derive a sequence for searching rooms. Then, based on visual observations and prior probabilities, we determine the locations of sub-goals. Through a path planning algorithm, the UAV progressively searches forward until the target is discovered. Experimental results demonstrate the effectiveness of the proposed method in addressing visual target navigation problems for UAV.