The integration of Large Language Models (LLMs) in Social Robots (SRs) has significantly enhanced SR capabilities and autonomy. SRs are known for their ability to engage in verbal and nonverbal communication with humans. Incorporating persuasive traits in SRs has been shown to improve human-robot interactions (HRI). However, research has not yet investigated the persuasive behavior in SRs powered by LLMs.

This thesis explores the impact of various robot knowledge prompts on the persuasive behavior and effectiveness of LLM-powered SRs. First, qualitative content analysis was conducted to assess interactions and the effects of different prompts. In a second step, a survey was conducted where participants rated the robot's behavior and hypothetical persuasiveness across different scenarios. Furthermore, the survey analyzed how knowledge prompts influenced perceived expressiveness and assertiveness, and how these factors affected hypothetical persuasion effectiveness.

The qualitative study found that both self-knowledge and target-knowledge had a noticeable impact on the robot's behavior, while context-knowledge had minimal influence on the SR's persuasive behavior. The survey revealed that self-knowledge and target-knowledge significantly enhance perceived assertiveness and expressiveness, while context-knowledge has no statistically significant effect. Moreover, assertiveness was found to have a strong impact on persuasion effectiveness, whereas expressiveness had a moderate influence. These findings highlight the importance of integrating knowledge prompts to enhance the persuasive behavior of LLM-powered SRs. This thesis provides first insights into the persuasive dynamics of LLM-powered SRs and suggests future directions to further advancing the field of HRI.