This study examines how biases in large language models (LLMs) influence public perceptions of the European Union (EU) in its Eastern neighborhood. Through three studies, we explore how EU public diplomacy shapes cultural biases in LLMs: (1) analyzing how LLMs incorporate EU-related public diplomacy from open-source data and tools like ChatGPT, (2) assessing variations in EU representation across different Eastern European languages, and (3) estimating the impact of introducing EU public diplomacy material into LLM training data. Through these three studies, the paper offers an innovative approach to the study of the EU’s external image by combining computational analysis with studies of public diplomacy and perceptions of international organisations. Overall, our analysis highlights how LLMs, which are increasingly used by media professionals and individual citizens for information purposes, can shape the EU's image abroad and thus its reputation and influence.