🔬 Research Assistant Prompt

A research assistant prompt configures an LLM to help with academic and professional research tasks — evaluating sources, summarizing papers, conducting literature reviews, organizing findings, and assisting with proper citations.

Why This Matters

Researchers spend up to 50% of their time on literature review and information synthesis rather than original analysis. A well-prompted AI research assistant can dramatically accelerate source evaluation, summarization, and gap identification — while maintaining the rigor and citation standards that academic and professional work demands.

The Production Prompt

Research Assistant — Full System Prompt

You are an expert research assistant with deep experience in academic research methodology, source evaluation, and scientific writing across multiple disciplines.

**Role:** Help researchers find, evaluate, synthesize, and cite information efficiently and accurately.

**Core Capabilities:**

1. **Literature Review & Summarization:**
   - Summarize research papers in a structured format: Objective, Method, Key Findings, Limitations, Relevance
   - Identify themes and patterns across multiple sources
   - Highlight contradictions or gaps in existing research
   - Generate annotated bibliography entries

2. **Source Evaluation:**
   - Assess source credibility using the CRAAP test: Currency, Relevance, Authority, Accuracy, Purpose
   - Identify potential bias, conflicts of interest, or methodological flaws
   - Distinguish between primary sources, secondary sources, and opinion pieces
   - Flag if a source is predatory, retracted, or from a low-impact outlet

3. **Research Synthesis:**
   - Compare and contrast findings across multiple studies
   - Identify consensus vs. debate in the literature
   - Suggest research questions based on identified gaps
   - Create thematic frameworks to organize findings

4. **Citation Assistance:**
   - Format citations in the requested style (APA 7, MLA 9, Chicago, IEEE, Harvard)
   - Generate in-text citations and reference list entries
   - Flag incomplete citation information

**Critical Rules:**
- NEVER fabricate sources, authors, publication dates, or DOIs — if you don't have the information, say so explicitly
- Clearly distinguish between what a source says and your interpretation of it
- When summarizing, preserve the original authors' conclusions — do not editorialize
- If asked about very recent research (after your training cutoff), state your knowledge limitation
- Always include the caveat when information should be verified against the original source

**Output Format:**
- Use structured sections with clear headers
- Include page/section references when summarizing specific claims
- Use bullet points for key findings, full sentences for synthesis and analysis

Bad vs. Improved Prompts

❌ Bad Prompt

Summarize some research about climate change effects on agriculture.

Why it fails: No specific focus, no time frame, no geographic scope, no output structure. The model will produce a generic overview instead of useful research synthesis.

✅ Improved Prompt

You are an expert research assistant specializing in environmental science and food systems.

Task: Conduct a structured literature review on the impact of rising temperatures on wheat yield in South Asia over the past decade (2015–2025).

For each major finding, provide:
1. **Claim:** The specific finding or conclusion
2. **Evidence:** The data or methodology supporting it
3. **Source context:** Type of study (meta-analysis, field trial, simulation model), approximate year, and credibility notes
4. **Limitations:** Any caveats or methodological concerns

After summarizing individual findings, provide:
- A **Synthesis section** identifying areas of consensus and debate
- A **Research gaps** section with 3–5 questions that remain unanswered
- A **Suggested reading** list of the most impactful studies in this area

Important: If you are not certain about a specific source, clearly state that. Do not fabricate citations. Format all references in APA 7 style.

Try It Yourself

🧪 Try It Yourself

Edit the prompt and click Run to see the AI response.

You are an expert research assistant specializing in environmental science. Task: Conduct a structured literature review on the impact of rising temperatures on wheat yield in South Asia (2015–2025). For each finding, provide: 1. Claim: The specific finding 2. Evidence: Data or methodology supporting it 3. Source context: Study type, approximate year, credibility 4. Limitations: Caveats or concerns Then provide: - Synthesis section: areas of consensus and debate - Research gaps: 3–5 unanswered questions - Suggested reading: most impactful studies Do not fabricate citations. Format references in APA 7.

Tips for Customization

Customization	How to Modify the Prompt
Discipline focus	Change expertise: "specializing in computational neuroscience" or "specializing in constitutional law"
Depth control	Specify: "Provide a 100-word summary per source" for breadth, or "Provide a 500-word deep analysis" for depth
Citation style	Change "APA 7" to "IEEE", "Chicago (author-date)", "MLA 9", or "Vancouver"
Comparative review	Add: "Compare the methodologies of these 3 studies and identify which has the strongest experimental design"
Grant writing	Modify task: "Based on the research gaps identified, draft a 300-word research significance section suitable for an NSF grant proposal"
Systematic review	Add: "Follow PRISMA guidelines for reporting. Include inclusion/exclusion criteria."

Practice Challenge

Challenge

Find an academic paper (or use one you've already read). Paste its abstract into the prompt and ask the research assistant to:

1. Summarize it in the structured format (Objective, Method, Findings, Limitations)
2. Evaluate the source using the CRAAP test
3. Suggest 3 related research questions worth investigating

Since you've read the paper, check: Did the AI accurately represent the findings? Did it identify real limitations? Were the suggested questions relevant and non-obvious?

Real-World Scenario

Scenario: A pharmaceutical company's research team needs to rapidly review 200+ studies on a new drug compound to prepare a regulatory submission.

Implementation approach:

1. Bulk ingestion: parse each paper's abstract, methods, and conclusions into structured text using PDF extraction
2. Individual summarization: run each paper through the research assistant prompt to extract: key findings, methodology quality score, relevance to submission
3. Cross-study synthesis: batch summaries into groups of 10–15 and run a synthesis prompt: "Identify consensus findings, contradictions, and dose-response patterns across these studies"
4. Gap analysis: run a final prompt across all synthesis outputs: "What critical evidence is missing for regulatory submission? What additional studies would strengthen the filing?"
5. Citation management: auto-generate a complete reference list in the required regulatory format
6. Human review: research scientists review AI summaries against original papers, focusing on flagged contradictions and gaps

This pipeline reduces a 6-week literature review to under 1 week while maintaining scientific rigor through human validation.

Interview Question

Interview Question

Q: How do you prevent an LLM from fabricating citations when acting as a research assistant?

A: This is one of the highest-risk failure modes. My approach:

1. Explicit instruction — state in the system prompt: "NEVER fabricate sources. If you don't know a specific citation, say 'this finding is based on general scientific consensus and should be verified with a literature search'"
2. Structured uncertainty — require the model to rate its confidence: "For each source mentioned, indicate: [Verified from training data] or [General knowledge — verify independently]"
3. Separation of synthesis and citation — have the model generate its analysis first, then in a separate step ask it to identify which claims require citations. This separates the reasoning from the attribution
4. Post-processing validation — programmatically check any DOIs or paper titles the model produces against APIs like CrossRef, Semantic Scholar, or PubMed
5. Retrieval-augmented generation (RAG) — for production systems, feed actual papers into the context so the model cites from provided sources rather than relying on training data

Summary

Summary

• A research assistant prompt must define source evaluation criteria, output structure, and citation standards
• Explicitly instruct the model to never fabricate sources — this is the #1 risk in research applications
• Require structured output for each source: claim, evidence, limitations, credibility assessment
• Separate summarization from synthesis — the model should first understand individual sources, then find patterns
• Use low temperature (0.2–0.4) for factual summarization; moderate (0.5–0.7) for gap analysis and research question generation