Can GPT separate sourced claims from its own inferences when comparing policy documents?

I’ve been using GPT to compare policy documents, and the summaries are useful, but I sometimes can’t tell which statements come directly from the text and which are the model’s interpretation.

Is there a reliable way to ask it to distinguish quotations or document-backed claims from its own synthesis? I’m especially interested in whether this holds up across several long documents, where a neat comparison can hide small differences in wording or scope.