Stage 2: COLLECT (15 minutes)
Rapidly gather 20-30 relevant sources using search strategies and citation snowballing
Speed Over Depth
At this stage, the goal is to cast a wide net based on titles and abstracts only. Deep reading comes later in Stage 3. Focus on volume and variety, not comprehension. Rapid triage is the key to efficient literature collection.
The 3-Step Collection Process
Step 2.1: Initial Search (5 minutes)
Use Google Scholar to search for keywords from the research question. Apply temporal and relevance filters, then open 10-15 promising papers based on titles and abstracts ONLY—do not read full papers yet.
Search Strategy Example:
"large language models" AND productivity AND ("software development" OR "content creation")Filters:
- Since 2022 (for recent research)
- Sort by relevance (for first pass)
Quick Evaluation Checklist (Prose Format):
The task at this stage is rapid triage. A paper passes the initial filter if the title directly addresses the research question, the abstract mentions empirical data or case studies, it is published in a peer-reviewed venue such as a journal or conference, and it has citable sources with real authors and institutions. If all four criteria are met, open the paper in a new tab for later processing. If any criterion is not met, skip the paper and move to the next result. This method ensures efficient use of the five-minute time budget.
Step 2.2: Citation Snowballing (5 minutes)
Select the 2-3 most relevant papers from the initial search and use them as "seed papers" to discover related work through forward and backward citations.
Forward Citations: Click "Cited by X" in Google Scholar to see newer papers that built on this work. These are valuable for seeing how the field has evolved.
Backward Citations: Look at the paper's References section to see foundational work it cites. These are valuable for understanding theoretical grounding.
Find 5-10 additional relevant papers this way. Pro tip: Use Connected Papers (connectedpapers.com) to visualize citation networks and discover related work that might have been missed with keyword search alone.
Step 2.3: Save to Zotero (5 minutes)
For each paper to include in the review, use the Zotero Connector browser extension to capture it with one click. Zotero automatically extracts title, authors, publication date, journal, DOI, and often downloads the PDF. Tag each paper with the project name (e.g., "LLM-productivity-review") for easy filtering later.
Expected Outcome: The collection should now contain 20-30 papers saved in Zotero, organized and ready for extraction.
Time Check: 15 minutes elapsed. No full papers have been read yet. This is intentional—the process optimizes for breadth before depth.
Why Citation Snowballing Works
Citation networks reveal connections that keyword search misses. Forward citations show impact and evolution of ideas over time. Backward citations show foundations and methodology precedents. Together, they ensure comprehensive coverage of the topic's literature landscape without requiring manual review of every possible search result.
Key Outcomes
By the end of Stage 2:
- 20-30 relevant papers saved in Zotero
- Papers tagged with project identifier
- Mix of recent work (forward citations) and foundational work (backward citations)
- PDFs automatically downloaded where available
- Zero full papers read (reading happens in Stage 3)
Next Stage: Stage 3: EXTRACT—systematic reading and data extraction from collected sources.