Conversational Interactions with NPCs in LLM-Driven Gaming: Guidelines from a Content Analysis of Player Feedback

The growing capability and availability of large language models (LLMs) have led to their adoption in a number of domains. One application domain that could prove fruitful is to video games, where LLMs could be used to provide conversational responses from non-playable characters (NPCs) that are more dynamic and diverse. Additionally, LLMs could allow players the autonomy to converse in open-ended conversations potentially improving player immersion and agency. However, due to their recent commercial popularity, the consequences (both negative and positive) of using LLMs in video games from a player’s perspective is currently unclear. On from this, we analyse player feedback to the use of LLM-driven NPC responses in a commercially available video game. We discuss findings and implications, and generate guidelines for designers incorporating LLMs into NPC dialogue.