Would you trust AI to help with systematic-review screening? Where should the boundary be?

I’m interested in how researchers think about AI-assisted literature screening and data extraction.

AI can potentially help prioritize records, surface patterns, and reduce repetitive work. But systematic reviews require transparency, reproducibility, and human judgment especially for include/exclude decisions and extracted study data.

Where would you find AI assistance genuinely useful?

• Prioritizing likely relevant titles and abstracts

• Suggesting exclusion reasons

• Highlighting PICO elements

• Extracting study characteristics

• Detecting duplicate or related publications

• Identifying missing data fields

• Summarizing extraction tables

And where should AI never act without explicit researcher review?

I’m building EvidenceFlow with the principle that researchers retain final control over screening, extraction, and analytical decisions. I’d value examples of what would make you trust or not trust this kind of workflow.