Large Language Models (LLMs) offer promising opportunities for enhancing pervasive games, particularly in refining game design, including its content, through the integration of Artificial Intelligence (AI). The goal of this work is to discuss the use of LLMs for Procedure Content Generation (PCG) in a pervasive game, as well as employing LLMs to evaluate the context accuracy of the generated content. For this, we developed a card game “Top Triumphs” inspired by Top Trumps, which features a PCG module composed of one AI system and three LLMs. The PCG process is started using the players’ geographic location. We also created CheckThis, a LLM-based tool that evaluates the contextual relevance and character accuracy of generated content, helping to tune the PCG process. Three experts evaluated the game content generated, confirming contextual and character accuracy with evaluation blocks reaching peaks of 92.85%. In the voting system part of CheckThis, one of the models managed to reach peaks of 83.3% acceptance. LLMs can significantly enhance pervasive game development by enabling more efficient design and testing workflows. Broader deployment requires additional validation, expansion, and ethical review.