Skip to content
Research Memex

Systematic Reviews

Building an SLR with Claude Code

Claude Code can help an SLR only when the project structure, prompts, and verification gates are explicit.

Agentic AI extends your reach across the literature, not your judgment of it.

Introduction: from cognitive overload to architectural oversight

Researchers conducting a Systematic Literature Review (SLR) often face cognitive overload. We identify, screen, and synthesize hundreds of articles, manage complex text datasets, and build arguments from existing literature. This process is time-consuming and can introduce inconsistency and error.

Claude Code can help when the workflow is explicit. If you use it haphazardly, it can lead to superficial work. If you use it strategically, it extends what a single researcher can inspect across the literature.

This guide demonstrates one approach: transforming your SLR project into an AI-navigable "codebase" where you act as the architect, directing the work. It assumes you have completed Session 4: Agentic Workflows and installed Claude Code.

Section 1: bootstrapping your SLR project

Step 1.1: launch and create project

Create your SLR project folder:

mkdir Systematic-Review-2025
cd Systematic-Review-2025
claude

Once Claude Code launches, it creates a CLAUDE.md file automatically. This is your SLR's "brain."

Step 1.2: the one-prompt project setup

Copy this entire prompt into Claude Code to bootstrap your SLR structure:

This is a systematic literature review project. Create the following directory structure:

- 01_search_and_screening/
- 02_data_extraction/
- 03_synthesis_and_writing/
- 04_manuscript_drafts/
- 00_literature_files/

Then populate CLAUDE.md with this SLR protocol:

# Systematic Literature Review Protocol

## Core Task
Systematic review to understand [YOUR RESEARCH QUESTION]

## File Structure
- /01_search_and_screening/ - Search strings, screening logs
- /02_data_extraction/ - Extraction tables (CSV/Markdown)
- /03_synthesis_and_writing/ - Theme analysis, drafts
- /04_manuscript_drafts/ - Final paper sections
- /00_literature_files/ - Full-text papers (markdown)

## Inclusion Criteria
- [YOUR CRITERIA - e.g., Peer-reviewed, 2020-2025, empirical]
- [Geographic/industry scope]
- [Methodology requirements]

## Exclusion Criteria
- [YOUR CRITERIA - e.g., Non-English, opinion pieces, duplicates]

## Citation Style
APA 7th Edition

## Custom Commands
- /screen: Evaluate abstract against inclusion criteria
- /extract: Extract data from full-text paper
- /critique: Review writing for quality and style

Claude will create the folders and set up your CLAUDE.md.

Step 1.3: organize literature files

Put your curated papers from Zotero into /00_literature_files/:

Naming convention:

smith2024a.md
jones2024b.md
lee2023.md

Use Better BibTeX citation keys, matching Zotero. See Zotero Setup Guide.

Reference papers in Claude:

"Analyze @/00_literature_files/smith2024a.md"
"Compare @/00_literature_files/smith2024a.md and @/00_literature_files/jones2024b.md"

Section 2: the three-tier agentic workflow

This framework helps you delegate SLR tasks so you can focus on the most important work.

Tier 1: automated assistant (mechanical tasks)

Goal: eliminate tedious, high-volume work.

Use cases:

Initial screening:

"Apply the /screen command to every abstract in
@/01_search_and_screening/database_export_abstracts.txt
and output results to screening_results.csv"

Bibliography management:

"Scan @/04_manuscript_drafts/full_draft.docx and generate
a complete, alphabetized bibliography in APA 7th edition"

Keyword counting:

"Search all files in @/02_data_extraction/ and count
occurrences of: 'scalability', 'scaling', 'growth'"

Tier 2: working partner (structured tasks)

Goal: accelerate structured content creation.

Use cases:

Data extraction:

"Run /extract on @/00_literature_files/smith2024a.md
and append results to @/02_data_extraction/extraction_table.md"

Drafting thematic sections:

"Using data in @/02_data_extraction/extraction_table.md,
draft a 'Emerging Themes' section. Synthesize findings across
studies and identify patterns."

PRISMA flowchart:

"Based on logs in @/01_search_and_screening/, create a PRISMA
flowchart showing: records identified, screened, included, excluded."

Task tracking:

"Create a todo list for synthesizing these 5 papers on organizational learning:
- Identify common themes
- Extract methodological approaches
- Map theoretical frameworks
- Identify research gaps
- Draft synthesis section"

Tier 3: Socratic sparring partner (deep work)

Goal: sharpen your theoretical contribution.

Use cases:

Identifying research gaps:

"Based on all papers in @/02_data_extraction/extraction_table.md,
what are the three most significant theoretical gaps? For each gap,
explain why it's important."

Note: see Coherence Fallacy for common failure modes.

Strengthening arguments:

"My main argument is [YOUR ARGUMENT]. Act as a skeptical reviewer
from [TARGET JOURNAL]. Formulate three powerful critiques based on
the evidence I've synthesized."

Future research avenues:

"Based on limitations across all papers, propose a research agenda
with three novel questions that address these gaps. Justify why each
represents a meaningful contribution."

Section 3: creating SLR-specific slash commands

3.1 screening command

Create .claude/commands/screen.md:

Evaluate this abstract against the inclusion criteria in @CLAUDE.md.

Abstract: $ARGUMENTS

## Screening Decision

Decision: [INCLUDE | EXCLUDE]

Reason: [Specific criterion met/failed]

Confidence: [High | Medium | Low]

Notes: [Any boundary cases or uncertainties]

Format as CSV row: decision, reason, confidence

Usage:

/screen This paper examines organizational scaling in tech startups
using survey data from 500 companies across Europe and Asia between
2020-2023...

3.2 data extraction command

Create .claude/commands/extract.md:

Extract structured data from this paper:

Paper: $ARGUMENTS

## Extraction Template

| Field | Content |
|-------|---------|
| Authors & Year | |
| Research Question | |
| Methodology | [Qual/Quant/Mixed] |
| Sample | [n=?, context] |
| Data Collection | |
| Analysis Approach | |
| Key Findings | [3-5 bullets] |
| Theoretical Contribution | |
| Limitations | |
| Future Research | |

Append this table to @/02_data_extraction/master_table.md

Usage:

/extract @/00_literature_files/smith2024a.md

3.3 critique command

Create .claude/commands/critique.md:

Review this writing section for quality:

Section: $ARGUMENTS

## Quality Checklist

1. Logical Flow: Are arguments well-structured?
2. Evidence Strength: Are claims supported by citations?
3. APA 7th Compliance: Citations formatted correctly?
4. Clarity: Is language precise and academic?
5. Coherence: Do paragraphs connect logically?

Provide specific feedback with line references.

Section 4: the complete SLR workflow

Phase 1: search & screening (Weeks 1-2)

Step 1: document search strategy

"Create a search log in @/01_search_and_screening/search_log.md
documenting:
- Databases used: Web of Science, Scopus, etc.
- Search strings and boolean operators
- Date ranges
- Total hits per database"

Step 2: screen abstracts

"Apply /screen to each abstract in
@/01_search_and_screening/abstracts.txt
Output to screening_results.csv with columns: ID, Decision, Reason"

Step 3: generate PRISMA counts

"Analyze screening_results.csv and generate PRISMA flowchart numbers:
- Records identified
- Records after deduplication
- Records screened
- Records excluded (with reasons)
- Full-text articles assessed
- Studies included in synthesis"

Phase 2: data extraction (Weeks 3-4)

Step 1: extract from each included paper

For each file in @/00_literature_files/:
/extract @/00_literature_files/[filename]

Step 2: compile extraction table

"Combine all extraction outputs into a single master table
at @/02_data_extraction/master_extraction_table.md"

Step 3: run quality check

"Review @/02_data_extraction/master_extraction_table.md
for missing data, inconsistencies, or errors. Flag any papers
needing manual review."

Phase 3: synthesis & analysis (Weeks 5-8)

Step 1: conduct thematic analysis

"Analyze @/02_data_extraction/master_extraction_table.md
and identify 5-7 major themes. For each theme, list:
- Papers that discuss it
- Key arguments/findings
- Points of agreement/disagreement"

Step 2: create methodological synthesis

"Create a methods summary table showing distribution of:
- Qualitative vs quantitative approaches
- Sample sizes and contexts
- Data collection methods
- Analysis techniques"

Step 3: perform gap analysis

"Based on all papers, identify:
- Methodological gaps
- Theoretical gaps
- Empirical/contextual gaps
- Temporal gaps

For each gap, explain significance and future research implications."

Phase 4: writing & refinement (Weeks 9-12)

Step 1: draft sections

"Using themes from @/03_synthesis_and_writing/themes.md,
draft a literature review section. Structure:
- Introduction to theme
- Discussion of each paper's contribution
- Synthesis of consensus and contradictions
- Implications for research"

Step 2: run quality control

/critique @/04_manuscript_drafts/literature_section_v1.md

Step 3: generate bibliography

"Extract all citations from @/04_manuscript_drafts/full_draft.md
and generate an APA 7th formatted bibliography"

Section 5: quality control & verification

5.1 common failure modes in SLR

Reference the Failure Museum for:

  • Citation confusion: AI inventing plausible citations
  • Context stripping: Missing nuances from papers
  • Coherence fallacy: Creating false consensus

5.2 verification checklist

After each AI-generated output:

  • Verify all citations exist in your Zotero library
  • Cross-check extracted data against original papers
  • Review for logical consistency
  • Check for hallucinated claims
  • Validate statistical information

5.3 human judgment gates

Critical checkpoints requiring human review:

  1. Inclusion decisions (borderline cases)
  2. Theme identification (theoretical framing)
  3. Gap analysis (novelty assessment)
  4. Theoretical contribution (intellectual merit)
  5. Final argument (coherence and originality)

Section 6: advanced SLR techniques

6.1 iterative screening

First pass - conservative:

/screen [abstract]
Decision: UNCERTAIN
→ Add to manual review queue

Second pass - human refinement: Review uncertain cases personally, then update screening criteria in CLAUDE.md

6.2 forward/backward citation tracking

Using Research Rabbit exports:

"Compare papers in @/00_literature_files/ with the citation network
exported from Research Rabbit. Identify:
- Highly cited papers we might have missed
- Citation clusters suggesting themes
- Outlier papers worth investigating"

6.3 meta-analysis preparation

Extract quantitative data:

"From all papers in @/00_literature_files/ with quantitative results,
extract effect sizes, sample sizes, and statistical significance.
Format as CSV for meta-analysis."

Section 7: parallel agent workflows

7.1 multi-agent paper analysis

Analyze from multiple perspectives:

"Use 3 parallel agents to evaluate @/00_literature_files/smith2024a.md:
- Agent 1: Assess methodology rigor
- Agent 2: Evaluate theoretical contribution
- Agent 3: Identify practical implications

Synthesize their assessments into a comprehensive review."

7.2 parallel screening

Screen batches simultaneously:

"Split the abstracts in @/01_search_and_screening/batch1.txt
into 3 groups. Use parallel agents to screen each group
with /screen command. Compile results when complete."

Section 8: export & deliverables

8.1 PRISMA flowchart

"Generate a complete PRISMA 2020 flowchart based on:
- Search log: @/01_search_and_screening/search_log.md
- Screening results: @/01_search_and_screening/screening_results.csv
- Inclusion log: @/01_search_and_screening/included_papers.md

Output as markdown table and mermaid diagram."

8.2 extraction table export

"Convert @/02_data_extraction/master_extraction_table.md
to CSV format for:
1. Import into Excel/R for analysis
2. Appendix for manuscript submission
3. Supplementary materials"

8.3 manuscript draft compilation

"Compile these sections into a complete draft:
- @/04_manuscript_drafts/01_introduction.md
- @/04_manuscript_drafts/02_methods.md
- @/04_manuscript_drafts/03_results.md
- @/04_manuscript_drafts/04_discussion.md

Add section numbers, format headers, generate bibliography."

Export to Word for final editing:

Use Zettlr for final formatting and citation management with @citekey references.

Section 9: ethical guidelines & academic integrity

9.1 transparency requirements

Document in your methods section:

### AI-Assisted Analysis

Screening and data extraction were assisted by Claude Code (Anthropic, 2025)
using standardized prompts defined in the review protocol. All decisions were
verified by [AUTHOR NAME]. AI outputs were subject to human review and
quality control checkpoints. The complete protocol and prompts are available
in the supplementary materials.

9.2 what to automate vs control

Automate (Tier 1):

  • Initial abstract screening (with human verification)
  • Basic data extraction
  • Bibliography formatting
  • Keyword counting

Work with (Tier 2):

  • Thematic coding
  • Methodology comparison
  • Drafting synthesis sections

Human Only (Tier 3):

  • Final inclusion decisions (borderline cases)
  • Theoretical framing
  • Novel gap identification
  • Argument construction
  • Quality assessment of contributions

9.3 verification protocol

For every AI-assisted task:

  1. Document the prompt used
  2. Review outputs for hallucinations
  3. Verify citations against Zotero library
  4. Check logical consistency
  5. Maintain audit trail

Section 10: troubleshooting SLR workflows

Screening inconsistencies

Problem: Different screening decisions for similar abstracts

Solution:

  • Refine inclusion criteria in CLAUDE.md
  • Add edge case examples
  • Create /screen_uncertain command for borderline cases
  • Maintain manual review log

Extraction errors

Problem: Missing or incorrect data in extraction table

Solution:

  • Verify against original papers
  • Update /extract template with more specific instructions
  • Add validation checks: "Review extraction for [paper] against original"

Citation hallucinations

Problem: AI invents citations that don't exist

Solution:

Section 11: example: complete screening workflow

Step 1: Prepare abstracts file

# Export from database to text file
# Format: One abstract per line, with ID prefix

Step 2: Screen abstracts in batches

"Process @/01_search_and_screening/web_of_science_export.txt:

For each abstract:
1. Apply /screen command
2. Log decision with ID
3. Track exclusion reasons
4. Flag uncertain cases for manual review

Output to screening_results.csv with columns:
ID, Title, Authors, Year, Decision, Reason, Confidence"

Step 3: Review uncertain cases

"Show me all abstracts marked 'UNCERTAIN' in screening_results.csv
for manual review"

Step 4: Generate screening stats

"Analyze screening_results.csv:
- Total abstracts screened
- Inclusion rate (%)
- Top 3 exclusion reasons
- Uncertain cases requiring manual review

Format as summary table."

Section 12: integration with Research Memex tools

12.1 Zotero → Claude Code pipeline

Workflow:

  1. Collect papers in Zotero (see Zotero Guide)
  2. Export to markdown via OCR (see OCR Guide)
  3. Move to /00_literature_files/
  4. Process with Claude Code

12.2 Claude Code → Obsidian pipeline

Export synthesis for knowledge base:

  1. Complete analysis in Claude Code
  2. Export synthesis notes to markdown
  3. Import to Obsidian vault
  4. Link with literature notes

See: Obsidian Setup Guide

12.3 Claude Code → Zettlr pipeline

Final manuscript preparation:

  1. Draft sections in Claude Code
  2. Export to /drafts/ folder
  3. Open in Zettlr for citation formatting
  4. Use @citekeys from Better BibTeX
  5. Export to Word/PDF for submission

See: Zettlr Setup Guide

Section 13: case study: scaling literature SLR

Example prompts from the course:

Initial setup:

"This SLR examines organizational scaling and scalability.
Inclusion: Peer-reviewed, 2020-2025, empirical studies on
business scaling. Exclusion: Non-English, opinion pieces,
case studies <3 companies."

Screening sample:

/screen @Coviello et al. 2024. This paper distinguishes
organizational scaling (growth in size) from scalability
(capacity to grow) using multi-method approach with
50 high-growth firms...

→ INCLUDE (meets all criteria, novel conceptual distinction)

Extraction sample:

/extract @/00_literature_files/coviello2024.md

→ Generated structured table with RQ, methods, findings, contribution

Synthesis:

"Analyze all papers with 'scalability' construct.
Compare definitions across studies. Identify conceptual confusion
vs. conceptual clarity. Draft 1-page synthesis."

Checklist: SLR with Claude Code

By the end of this workflow, you should have:

  • Created SLR project structure (5 folders)
  • Written CLAUDE.md with SLR protocol
  • Created /screen, /extract, /critique slash commands
  • Screened abstracts with quality verification
  • Extracted data into master table
  • Conducted thematic synthesis
  • Generated PRISMA flowchart
  • Drafted literature review sections
  • Verified all AI outputs against original papers
  • Documented AI use for methods section
  • Maintained audit trail of prompts

Resources

SLR methodology

Research Memex integration

Tool integration

cite this page

Lin, X. (2026). Building an SLR with Claude Code. Research Memex. https://research-memex.org/docs/case-studies/systematic-reviews/claude-code-slr-workflow

@misc{docs-case-studies-systematic-reviews-claude-code-slr-workflow-2026,
  author = {Xule Lin},
  title = {Building an SLR with Claude Code},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/case-studies/systematic-reviews/claude-code-slr-workflow}},
  note = {ORCID: 0000-0001-7885-4194}
}

one renderingthe source remains