AI for literature review: a practical workflow

I've been refining a workflow for research where I feed each paper into an AI to generate a structured summary: core question, methods, key findings, and limitations. Then I compile all these into a matrix to map the field. It saves hours, but I worry about missing nuance. Curious how others handle this in their research workflows. Do you use AI for screening or synthesis? What has worked, and where has it failed you? I'm also exploring whether this can be integrated with reference managers like Zotero.