Research Workflow
Design a systematic research workflow from discovery to output. Learn capture, organization, synthesis, and writing stages with practical tool recommendations.
Most people research by accident.
They Google something.
They read a few pages.
They remember (or forget) what they read.
They write something.
This is not a workflow. It's chaos.
Good research is reproducible. Documented. Intentional.
A documented research workflow ensures you can:
This guide covers building a research workflow from scratch.
You find a useful paper.
You think "I'll remember this."
Three weeks later, you forget where it came from.
Result: Lost sources. Panic during citation.
You have 200 bookmarks and 50 files.
They're named "Research1", "Article", "Stuff".
You can't find anything.
Result: Duplicated research. Wasted time.
You find conflicting sources.
You don't know which to trust.
You cite the one you read first.
Result: Weak research. Poor decisions.
You gather 50 sources.
You have no system for comparing them.
You write from memory.
Result: Incomplete coverage. Missed nuances.
Solution: A documented workflow that forces intentionality at each stage.
Goal: Find sources and decide if they're worth investigating.
Process:
Outcome: A list of candidate sources
Goal: Save the full text and metadata of promising sources.
Process:
Outcome: Organized archive with searchable full text
Goal: Assess credibility and relevance of each source.
Process:
Outcome: Curated sources with credibility assessment
Goal: Extract key insights and connect them.
Process:
Outcome: Synthesized knowledge ready for writing
Goal: Turn research into finished work.
Process:
Outcome: Finished article/paper with citations
Vague: "Tell me about AI"
Clear: "What are the ethical implications of AI in criminal justice systems?"
Specificity drives search strategy and determines what counts as "relevant."
| Source | Best For | Drawback |
|---|---|---|
| Google Scholar | Broad, multidisciplinary | Less curated |
| PubMed | Medical/health research | Limited to health |
| JSTOR | Academic depth | Requires subscription |
| Fast, broad | Noisier results | |
| Personal networks | Expert perspectives | Subjective |
Search each source with identical search terms (reproducibility).
Create a search log:
Date: 2025-01-15
Question: "Ethical implications of AI in criminal justice"
Search term: "AI criminal justice ethics"
Source: Google Scholar
Results: 1,240
Skimmed: 50 titles
Promising: 12
This allows someone else to repeat your search.
Read title + abstract. Decide quickly:
Don't read full text yet. You're filtering, not analyzing.
Export filtered sources to your citation manager.
You now have a prioritized list to dive into.
Citation Manager (required):
Web Clipper (optional but recommended):
Import using:
Check each source for complete data:
Fix incomplete metadata now (it's harder later).
Create a tag structure (don't create folders; tags are more flexible):
Topic tags: #ai, #criminal-justice, #ethics
Status tags: #read, #toread, #skimmed
Quality tags: #highquality, #secondary, #opinion
Tag each source immediately after adding.
Add a note to each source:
"Why I saved this: Discusses algorithmic bias in risk assessment tools"
Later, you'll remember context.
For each source, evaluate:
Author credibility:
Methodology:
Recency:
Bias:
For each source, document:
Source: Smith, J. (2023). "AI in criminal justice"
Credibility: HIGH
- Published in peer-reviewed journal
- Author is criminal justice scholar with 15 years experience
- Cites other high-quality sources
Bias: LOW-MEDIUM
- Generally balanced
- Acknowledges limitations
- Doesn't overstate findings
Relevance: HIGH
- Directly discusses ethical implications
- Focuses on prediction bias (my key interest)
Key findings:
- AI risk assessment tools have 20-30% higher error rates for minority defendants
- Error compounding across system stages
- Regulatory frameworks still inadequate
Conflicts/gaps:
- Doesn't address bias mitigation strategies
- Limited sample size (3 jurisdictions)
Create a comparison matrix:
| Source | Method | Finding on Bias | Finding on Regulation | Quality |
|---|---|---|---|---|
| Smith 2023 | Study | 20-30% error | Insufficient | High |
| Jones 2022 | Opinion | Not quantified | Absent | Medium |
| Brown 2024 | Study | 15-25% error | Emerging | High |
This shows where consensus exists and where sources conflict.
When sources disagree:
Document the disagreement and investigate:
Conflict: Smith claims AI bias is 20-30%. Jones claims <5%.
Investigation:
- Smith studies U.S. systems
- Jones studies U.K. systems
- Different regulatory environments explain difference
- Conclusion: Both correct in their contexts
This prevents you from making false conclusions.
For each source, pull out:
Example:
Key finding: AI risk assessment tools show 20-30% higher error rates for minority defendants
Supporting evidence:
- Smith et al. analyzed 10,000 cases across 3 U.S. states
- Controlled for legal factors (prior record, charge severity)
- Racial bias persisted even after accounting for legal factors
Implications:
- Current tools perpetuate criminal justice inequities
- Risk assessment alone isn't sufficient; humans need override power
Limitations:
- Study limited to 3 jurisdictions
- Doesn't test bias mitigation strategies
- Doesn't address why bias exists
Read all your synthesis notes.
What patterns emerge?
Theme 1: Technical bias in AI
- Multiple sources document algorithmic bias
- Causes: biased training data, design decisions
Theme 2: Regulatory gaps
- Current regulations inadequate
- Need sector-specific oversight
Theme 3: Mitigation strategies
- Few sources discuss solutions
- Research gap identified
Write comparative notes:
# AI Bias: Sources Compare
All sources agree on:
- Bias exists in current AI criminal justice tools
- Root cause is biased training data
Sources disagree on:
- Magnitude (Smith: 20-30% error vs Jones: <5%)
- Regulatory solution (Brown: stronger oversight vs Wang: industry self-regulation)
Research gap:
- Limited discussion of how to audit and mitigate bias
- No sources address long-term justice system impact
This becomes your article outline.
Create an outline based on themes:
1. Introduction: Define the problem
2. Technical bias exists (theme 1 + evidence)
3. Current regulations are inadequate (theme 2 + evidence)
4. Solutions are emerging (theme 3 + evidence)
5. Research gaps remain (what we don't know)
6. Conclusion: Call to action
Draft each section using your synthesis notes.
Your citation manager generates citations automatically (most managers integrate with Word/Google Docs).
Every claim should be supported:
"AI risk assessment tools show 20-30% higher error rates for minority defendants [Smith, 2023]."
This prevents accidental plagiarism and strengthens your work.
Before finalizing:
If you've missed major sources, your research process worked (you identified the gap).
What you searched, where, when, how many results
All sources, with reason for inclusion/exclusion
Credibility ratings and reasoning for each source
Key findings, conflicts, themes, gaps
Why you made major decisions (included/excluded source, chose interpretation, etc.)
Why document?
| Stage | Best Tool | Alternative |
|---|---|---|
| Discovery | Google Scholar | PubMed, JSTOR |
| Capture | Zotero | Mendeley |
| Organization | Citation Manager tags | Folders (less recommended) |
| Synthesis | Note app (Obsidian, Notion) | Citation manager notes |
| Output | Word/Google Docs | Overleaf (for LaTeX) |
This feels like extra work initially, but it saves time in writing (you're organized).
Workflow pays for itself after your second article.
You cite a blog post with equal weight as a peer-reviewed study.
Fix: Evaluate and differentiate source quality.
You save 100 sources without reading them.
Later you cite blindly.
Fix: Read abstracts during discovery. Don't save everything.
You repeat searches. Different results. Confusion.
Fix: Document every search with date, terms, results.
You save sources with incomplete citations.
During writing, you can't cite properly.
Fix: Check metadata immediately after capturing.
A documented research workflow transforms ad-hoc searching into reliable, reproducible research.
Five stages:
Benefits:
Start this week:
By next week, you'll have a system that becomes more valuable the more you use it.
For more on research, see Systematic Literature Review. For citations, check Citation Management.
Research systematically. Document intentionally. Publish confidently.
More WebSnips articles that pair well with this topic.
Build a bulletproof citation management system. Compare Zotero and Mendeley, master metadata, and integrate with your writing workflow.
Compare the best research tools for 2025 across every category — web clipping, citation management, literature search, AI research, and synthesis tools.
Learn when to skim vs read deeply in your research workflow. A decision framework for strategic reading that maximizes insight per hour invested.