Even if that specific document wasn't in the training data, there could be many similar documents from others at the time.