Why don't LLMs stick to single source unless provide them the pdf of that source?

So I've seen many youtube channels and some article claiming that most of llm models have run out of text to train their models. So basically that means that LLMs have been trained on most fo the publically available books and journals. So when I ask any LLM