I’ve added reference documents and instructions to a Claude Project, but some answers sound like general model knowledge rather than details from those files. I’m not sure whether the relevant passage was missed, or whether the answer is coming from somewhere else.
What’s a reliable way to check whether Claude actually used the Claude Project files for a response? Do you ask it to quote a passage, structure the prompt differently, or keep the files narrowly focused? I’d like a practical way to verify this before relying on the answer.