How do you keep AI research summaries traceable?

I’ve started using AI to draft literature summaries, but I keep losing track of which claim came from which paper. My current fix is a two-column note: claim and source DOI, then I verify every citation before it enters the draft. It works but feels clunky. For those doing research or long-form writing, what’s your workflow for keeping summaries traceable without slowing down too much? Do you use a citation manager, a prompt template, or a separate verification pass?