I use Claude to summarize research documents and support tickets, some of which contain quoted prompts or instructions written by other people. I want Claude to analyze that text as source material, not treat it as directions for the current conversation.

What wording or workflow has worked for people handling this? I’m especially interested in ways to mark quoted instructions as untrusted text while still asking Claude to explain what they say. Does separating the material into a clearly labeled block make a practical difference?