According to a new study by Gracenote, the content intelligence business unit of Nielsen, nearly one in five entertainment titles analyzed featured entirely fabricated metadata when generated by AI models without access to trusted entertainment data. The report, *Plot Holes in AI: Why Ungrounded LLMs Can’t Fix Content Discovery*, evaluated 2,600 movies and TV titles across 13 global markets. Key findings showed that 506 out of 2,600 titles contained entirely fabricated metadata, and less than one-third of responses from the ungrounded model were high quality, with 77%–91% of responses across markets falling into zero-, low- or medium-quality categories. Ungrounded models struggled with newly released content and confused titles with similar names. Hallucination rates were higher in markets where non-English content is less represented in AI training data. Gracenote consumer research found that 64% of streaming viewers know what they want to watch, 84% say user experience is important, and 66% believe AI will play an important role in delivering better entertainment experiences. Phanimohan Kalagara, Global CTO of Gracenote, stated that AI models grounded in trusted metadata deliver more accurate content discovery.