I think the ability to rate individual resources used in responses, similar to the Good Prompt/Bad Prompt feedback would be really awesome. This could help identify sources that are outdated, irrelevant, or especially useful. Over time, aggregated feedback could improve the quality and relevance of the resources cited in future responses while helping surface more trustworthy sources.
The feedback for each source should allow the user to be able to specify the quality of the source and whether the model interpreted it correctly.