Systematic review support, and its limits
Lesson 3 of 6 · Literature at scale
1 · Learn the move · Constraints + negative instructions
A systematic review is a protocol-governed instrument - registered criteria, documented screening, two independent human screeners - and a model can help without touching what makes it systematic. The boundary is drawn with negative instructions. The model MAY: translate search strings across databases, deduplicate results, apply YOUR registered criteria as a first-pass screen, every exclusion tagged and queued for human confirmation, and check your reporting against a checklist. The model MAY NOT: make final include/exclude calls, stand in as the second reviewer - screener independence is the design; a model you steer is not independent - pool effect sizes, or assess study quality. And before any of it: the target venue's and registry's current policy on AI assistance in evidence synthesis is verified and disclosed, not assumed - if unclear, the question routes to the editor, not to hope.
Support my systematic review under these constraints. You MAY: translate my search string for [databases]; deduplicate the export; first-pass screen against my registered criteria - [paste criteria verbatim] - tagging every exclusion with its criterion number and marking ALL decisions PROVISIONAL - HUMAN CONFIRMATION REQUIRED; log counts at every stage for the flow diagram. You MAY NOT: finalize any include/exclude, act as second screener, pool effects, or score study quality. Every output states its provisional status in the header. Screening record format: ID / DECISION (provisional) / CRITERION / the abstract phrase that triggered it.
2 · Your turn. You write the prompt
Your scoping review on biomarker validation methods has 1,240 deduplicated records and registered criteria, and your co-reviewer starts screening next week. Set the model up as the first-pass sorter whose every call lands in a human queue - and nothing more than that.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.