We formalise quoting in conversation, release a training set and benchmark, and introduce a tiny adapter that lets LLMs exploit quoted spans with zero prompt overhead.