Stage 3: EXTRACT (20 minutes)
Pull key information from sources using AI while making critical judgments
The Division of Labor
AI extracts information at scale—summarizing papers, identifying patterns, and organizing findings. However, YOU evaluate quality, credibility, and relevance. This is where human judgment is most critical. The AI can process dozens of papers in minutes, but only a researcher with domain knowledge can determine which findings are trustworthy and which are central to the research question.
Extraction Workflow
Pull key information from all collected sources using AI assistance, while making critical judgments about methodology, credibility, and relevance.
Step 3.1: Set Up Your AI Workspace (2 minutes)
Configure your AI assistant with a custom instruction that defines its role as a research assistant for systematic literature review.
For Claude: Create a new Project called "Research: [Your Topic]" and set this custom instruction:
You are a research assistant helping with a systematic literature review.
Research Question: [YOUR QUESTION]
Your role:
- Summarize papers accurately and concisely
- Extract key findings, methods, and limitations
- Identify themes and patterns across papers
- Flag contradictions or inconsistencies
- Always cite sources when making claims
Do NOT:
- Invent citations or data
- Speculate beyond what papers state
- Provide opinions without evidenceFor ChatGPT or Gemini: Start a new conversation and paste the same instruction as your first message.
Step 3.2: Bulk Processing (15 minutes)
Process collected papers in batches of 5-7 to maintain AI context while avoiding overload.
For each batch:
Upload PDFs if your AI supports it (Claude Projects and ChatGPT Plus both handle PDFs). If not, copy-paste the abstract, introduction, and conclusions from each paper.
Use this extraction prompt:
I've uploaded [NUMBER] papers. For each paper, extract:
1. **Citation** (Author, Year, Title)
2. **Research Question** - What did they investigate?
3. **Methods** - How did they study it? (sample size, approach, data)
4. **Key Findings** - What did they discover? (specific numbers/results)
5. **Limitations** - What are the caveats?
6. **Relevance** - How does this relate to my research question?
Format as a structured table for easy comparison.Critical Review (YOUR JOB): As the AI generates summaries, you must evaluate each paper on three dimensions. First, assess methodology: Does the approach make sense for the research question? Is the sample size adequate? Are the measures valid and reliable? Second, evaluate source credibility: Who funded the research? Are there potential conflicts of interest? Is this published in a reputable venue? Third, determine true relevance: Does this paper directly answer your research question, or is it tangentially related? Does it provide evidence that strengthens or challenges your working hypothesis?
Based on your evaluation, mark each paper with one of four categories: Core (directly answers your question with solid evidence), Supporting (provides context or partial evidence), Tangential (interesting but not central to your question), or Exclude (doesn't actually fit your research scope).
Repeat for all papers. By the end, you should have structured summaries of 20-30 papers, categorized by relevance, with key findings extracted.
Time Check: 35 minutes elapsed since Stage 1 began.
Critical Review: Your Bottleneck Responsibility
This evaluation step is where YOU are the bottleneck—AI cannot perform these judgment calls. Only a researcher with domain knowledge can determine whether a methodology is appropriate for its research question, whether a source is credible given potential conflicts of interest, and whether a paper truly addresses the specific research question at hand. These are not mechanical tasks that can be delegated to AI. This is where your expertise matters most.
Expected Outcome
Structured summaries of 20-30 papers, each categorized by relevance tier (Core, Supporting, Tangential, Exclude). The extraction table format enables easy comparison across studies, allowing you to identify patterns, contradictions, and gaps. You are now ready to synthesize findings and identify emergent themes across the literature.