How to Build a Paper Search Strategy
Daniel Ha · Seoul National University PhD StudentBreak down your research question using the PICO framework, then secure 5 to
10 core papers with AI semantic search using natural language. Expand your
collection through forward and backward citation tracking, and refine results
with year, citation count, and access filters. Use a PRISMA flow diagram to
document your selection process systematically.
Why Do You Need a Search Strategy?
With a search strategy, you can find the papers you need quickly, feel confident that nothing important was missed, and demonstrate the rigor of your literature review. "Can't you just type keywords into Google Scholar?" Sure, that will return results. But keyword-based searching has fundamental limitations.
- It misses papers that express the same concept with different terms ("remote instruction" vs. "online learning" vs. "e-learning")
- You waste time scrolling through 3,000 results one by one
- If you do not know the exact keywords, it is hard to even begin
With a search strategy, you can find relevant papers efficiently, gain confidence that your coverage is comprehensive, and demonstrate the rigor of your literature review to reviewers.
Step 1: Structure Your Research Question for Search
The starting point for searching is not keywords but your research question. Use the PICO/PEO framework to break your question into components.
| Element | Description | Example (Education Research) |
|---|---|---|
| P (Population) | Study population | University students, teachers |
| I (Intervention/Exposure) | Intervention or variable | AI tutoring, flipped learning |
| C (Comparison) | Comparison group | Traditional instruction, control group |
| O (Outcome) | Outcome variable | Academic achievement, learning motivation |
This decomposition naturally generates both natural language queries for AI semantic search and search terms for traditional keyword-based searching.
Step 2: Secure Core Papers with AI Semantic Search
Searching in Natural Language
Traditional search requires you to know the exact keywords before you start, but AI semantic search lets you enter your research question as-is.
| Search Method | Input Example | Characteristics |
|---|---|---|
| Keyword search | "AI tutoring" AND "academic performance" AND "higher education" | Requires exact terms, risk of missing synonyms |
| AI semantic search | "Effect of AI-based personalized learning on university students' grades" | Natural language input, meaning-based matching |
AI semantic search understands meaning, not just words. Even if the keyword "AI tutoring" is not included, it finds papers about similar concepts like "intelligent tutoring systems" or "adaptive educational technology."
Traditionally, paper searches required different databases for each field.
| Field | Major Databases |
|---|---|
| General | Google Scholar, Scopus, Web of Science |
| Medicine/Health | PubMed, MEDLINE, Cochrane Library |
| Education | ERIC, Education Source |
| Psychology | PsycINFO, PsycArticles |
| Business/Economics | SSRN, EconLit, Business Source Complete |
| Engineering/CS | IEEE Xplore, ACM Digital Library |
| Korean Academic | RISS, KCI, DBpia, KISS |
Nubint AI's AI Paper Search covers 280 million academic papers across these databases in a single search. Instead of searching each database separately, enter your research question in natural language and find relevant papers across all fields through meaning-based matching.
Secure 5-10 Core Papers First
Do not try to find every paper at once. The first goal is to secure 5-10 core papers in your field.
- Find review articles first — Systematic reviews and meta-analyses serve as a map of the field
- Highly cited papers from the last 3-5 years — Show the current direction of the field
- Papers directly related to your research question — Papers with overlapping PICO elements
Enter your research topic into the Literature Review agent, and the AI will analyze related papers and organize key research trends, major findings, and research gaps. By searching and analyzing simultaneously, you can grasp the initial landscape of the literature much faster.
Step 3: Expand Through Citation Tracking
Once you have secured 5-10 core papers, citation tracking is far more efficient than cycling through different keyword combinations.
Backward Citation
Check the reference lists of your core papers. Following the prior works they cited helps you quickly identify the foundational literature in the field.
Forward Citation
Check the subsequent studies that cited your core papers. Use Google Scholar's "Cited by" link or Web of Science's "Cited by" feature. This is effective for identifying the latest research trends.
Snowballing
Repeating backward and forward citation tracking causes related papers to snowball. When no new papers are discovered — that is the saturation point of your search.
Nubint AI's Research Flow Explorer agent automatically analyzes the citation network of a specific paper and shows key prior works and follow-up studies. It significantly reduces the time spent manually tracing reference lists one by one.
Step 4: Filter and Refine Search Results
Once you have gathered papers through AI semantic search and citation tracking, refine them with filters.
Narrowing Down Results
If search results in Nubint AI Paper Search are too broad, use the filter feature to narrow them down.
| Filter Criterion | How to Use |
|---|---|
| Publication year | Limit to the last 5-10 years; adjust by field |
| Citation count | Prioritize highly cited papers |
| Open access | Filter for full-text accessible papers |
| Paper type | Distinguish review articles, empirical studies, etc. |
Keeping a Search Log
In systematic reviews, reproducibility of the search process is required. Record each search session.
| Item | What to Record |
|---|---|
| Date | 2026-04-15 |
| Database | Scopus |
| Search query | ("AI tutoring" OR "intelligent tutoring") AND "higher education" |
| Filters | 2020-2026, English, Journal |
| Results | 247 hits |
| Selected | 32 selected based on title/abstract screening |
Step 5: Screen and Organize
Screening Criteria
- Year: Last 5-10 years as default; adjust by field
- Document type: Journal articles, dissertations, conference papers, etc.
- Language: Limit to languages you can read
- Title/abstract screening: Judge relevance to your research question
- Full-text review: Methodological quality, research design appropriateness
PRISMA Flow Diagram
In systematic reviews, the PRISMA flow diagram transparently shows the screening process.
| Stage | Description | Example |
|---|---|---|
| Identification | Total search results | 1,240 |
| Duplicate removal | Remove duplicates across DBs | → 890 |
| Title/abstract screening | First-round selection by relevance | → 120 |
| Full-text review | Apply inclusion/exclusion criteria | → 45 |
| Final inclusion | Papers used in analysis | → 32 |
Saving and Organizing Papers
Organize papers immediately after searching. If you put it off, you will lose track of sources.
- Reference managers: Zotero, Mendeley, EndNote
- Notes/tagging: Tag each paper with key findings, methodology, relevance
- Paper library: Saving papers to Nubint AI's Paper Library allows agents to reference them during analysis
Search Strategy Summary: A Three-Phase Approach
| Priority | Method | Purpose | Tools |
|---|---|---|---|
| 1st | AI semantic search | Quickly secure 5-10 core papers | Nubint AI Paper Search, Literature Review agent |
| 2nd | Citation tracking | Expand network from core papers | Research Flow Explorer agent |
| 3rd | Filtering & refinement | Narrow down and screen results | Year, citation count, type filters |
Following this sequence lets you start quickly even at the early stage when you do not know the right keywords, and keywords naturally accumulate as you read more papers.
Summary
The key to paper searching is not what keywords to enter, but what question you need the papers to answer. Secure core papers quickly with AI semantic search, expand the network through citation tracking, and supplement with keyword searches. Recording your search process allows you to demonstrate the rigor of your literature review.