As large language models (LLMs) impress with their content creation and natural language processing capabilities, media companies are racing to harness their potential – but is this adoption truly aligned with organisational needs? This paper uncovers socio-technical (mis)alignments in LLM adoption, highlighting the socio-technical tensions that arise, including task misfits, structural challenges, and employee resistance. Through an embedded multiple case study of six use cases in three German media companies, we reveal that while LLMs are powerful and accessible, they often fail to reach their full potential due to task misalignments because of complexity, cognition, volume, and context. We contribute to the artificial intelligence (AI) adoption literature with insights specific to generative AI in content-driven industries and help organisations better align LLM adoption.