A systematic review is a research method, not an automated summary of several papers. Standards vary by discipline and review type, but require a prespecified question, reproducible retrieval, transparent selection, and assessment of bias and heterogeneity. Automation does not transfer methodological responsibility.
When this workflow is useful
- Answering a bounded question systematically
- Producing auditable evidence for guidance or decisions
- Updating a review and identifying newly eligible studies
A practical sequence
- 01
Prespecify the protocol
Define the question, sources, limits, eligibility criteria, outcomes, extraction fields, and synthesis plan. Register where appropriate and follow relevant reporting guidance.
- 02
Run reproducible searches
Search multiple appropriate sources and retain raw exports and logs, including platform, date, complete query, and every filter.
- 03
Deduplicate and screen in stages
Screen titles and abstracts before full text. Machine ranking may prioritize records, but stopping rules, sampling checks, and human review must be explicit.
- 04
Extract, assess, and synthesize
Pilot a structured form, assess bias with a method suited to the design, and report flow, exclusions, heterogeneity, uncertainty, and protocol deviations.
Quality-control questions
- Were protocol and criteria fixed before results were known?
- Does automation retain human review and an audit trail?
- Are queries and full-text exclusions recorded?
- Does synthesis fit the designs, bias, and heterogeneity?