Ironically I used AI for this question, and what came out is 'Echo prompt' apparently it's a thing: repeating LLM instructions twice increases accuracy.

Another idea was Chain of Thought leakage. It's like as if a chef saying "Now I am adding salt because the customer requested so"