Literature Review & Discovery

Finding relevant papers isn’t the hard part — it’s finding all of them, understanding how they connect, and synthesizing what they collectively say. A manual literature search catches about 60-70% of relevant papers. AI-assisted search, combined with citation network mapping, pushes that toward 90%+.

The AI Literature Review Stack

Each tool serves a specific purpose. Using them together gives you coverage no single tool provides.

✅ Quick Check: You’re studying the effect of sleep on memory consolidation. Which tool would you use to find out how many papers support vs. challenge the “active consolidation” hypothesis? (Answer: Scite — its Smart Citations classify every citation as supporting, contrasting, or mentioning, giving you a quantitative picture of the evidence landscape.)

Structured Literature Search

A productive AI search starts with a structured query, not a vague topic.

Help me design a systematic literature search for:

Research question: [your specific question]
Key concepts: [2-4 main concepts]
Population/context: [who or what you're studying]
Timeframe: [last N years, or specific range]
Exclusions: [what you're NOT looking for]

Generate:
1. Primary search terms with synonyms and MeSH terms
2. Boolean search strings for PubMed, Scopus, and Web of Science
3. Inclusion/exclusion criteria for screening
4. A screening priority rubric (high/medium/low relevance signals)

Example output for “AI in surgical decision-making”:

• Primary: (“artificial intelligence” OR “machine learning” OR “deep learning”) AND (“surgical” OR “operative” OR “perioperative”) AND (“decision” OR “clinical judgment” OR “treatment planning”)
• MeSH: “Artificial Intelligence”[MeSH] AND “Decision Making, Computer-Assisted”[MeSH] AND “Surgical Procedures, Operative”[MeSH]

Building Citation Networks

Once you find 3-5 key papers, Connected Papers reveals the landscape around them.

How to use citation networks effectively:

1. Start with your most relevant paper — the one closest to your research question
2. Examine the graph — papers cluster by topic; dense connections mean related work
3. Check the “prior works” view — foundational papers your key paper builds on
4. Check the “derivative works” view — papers that built on your key paper
5. Repeat with a second key paper from a different cluster to catch adjacent fields

The papers that appear in multiple networks (connected to several of your key papers) are often the most important ones to read — and the ones keyword searches miss.

Evidence Synthesis with AI

After collecting your papers, AI helps you extract and compare data across studies.

I have [N] papers on [topic]. Help me create a synthesis matrix:

For each paper, extract:
- Study design (RCT, cohort, cross-sectional, etc.)
- Sample size and population
- Key intervention or variable
- Primary outcome measures
- Main findings (with effect sizes if reported)
- Limitations noted by authors
- Funding source

Then identify:
1. Points of agreement across studies
2. Contradictions or conflicting findings
3. Gaps — questions none of the studies address
4. Methodological trends (are newer studies using better designs?)

✅ Quick Check: You’ve built a synthesis matrix for 15 papers on exercise and depression. Twelve show significant effects, but three show no effect. Should you exclude the three negative studies? (Answer: Never. Selective reporting creates publication bias. Include all studies, examine why results differ — the three negative studies might have smaller samples, different exercise types, or shorter durations. Report the full evidence landscape.)

Managing Ongoing Literature Monitoring

Research doesn’t stop after your initial review. Use AI to stay current.

Research Rabbit workflow:

1. Create a collection with your 10-20 most important papers
2. Research Rabbit recommends related papers — including new publications
3. Review recommendations weekly (15 minutes)
4. Add relevant papers to your collection (the recommendations improve over time)

AI alert prompt:

Based on my research focus on [topic], generate:
1. Search alert queries for Google Scholar Alerts (2-3 queries)
2. Key authors to follow (based on my bibliography)
3. Key journals to monitor
4. Conference proceedings to check annually
5. Preprint servers to watch (bioRxiv, arXiv, SSRN as appropriate)

Practice Exercise

1. Take a current research question and run it through the structured search prompt above
2. Find 3 key papers using Semantic Scholar, then map their citation network in Connected Papers
3. Use Scite to check the citation sentiment for your most-cited reference — are papers supporting or challenging it?

Key Takeaways

• Layer multiple AI tools: Semantic Scholar for search, Connected Papers for networks, Elicit for extraction, Scite for citation quality
• Structured search queries with Boolean operators and MeSH terms outperform vague keyword searches
• Citation networks catch papers that keyword searches miss — always map from at least 2-3 key papers
• Evidence synthesis should include all studies, including negative results — selective reporting creates bias
• Set up ongoing monitoring with Research Rabbit and Google Scholar Alerts to stay current after your initial review

Up Next

In the next lesson, you’ll use AI to design better experiments — from hypothesis refinement and power analysis to protocol optimization and variable selection.