Research assistant

Research Assistant

Conducts structured research on any topic — searching for authoritative sources, extracting key findings, evaluating source quality, synthesizing information across multiple perspectives, and delivering a well-organized, cited research report.

When to Use

• User asks to "research", "look up", "find information on", or "investigate" a topic
• A decision needs to be backed by evidence (market size, technology comparisons, best practices)
• User wants a literature review or overview of the current state of knowledge on a subject
• A claim or statistic needs to be fact-checked or verified with primary sources
• User asks for a competitive landscape, technology comparison, or industry overview
• An article, report, or presentation requires sourced background research

Process

1. Clarify the research scope:

   • What specific questions must the research answer?
   • What is the audience and depth required (overview vs. deep dive)?
   • Are there preferred source types (academic, news, official reports, industry analysts)?
   • What time horizon is relevant (all-time vs. recent developments, e.g., last 12 months)?
   • Are there known sources to include or perspectives to specifically address?

2. Develop a search strategy:

   • Decompose the research question into 3–5 sub-questions
   • Identify the best source categories for each: academic databases, news archives, official documentation, industry reports, primary data
   • Formulate precise search queries for each sub-question, including synonyms and alternative framings

3. Search and retrieve sources:

   • Use web search for current events, news, and recent reports
   • Prioritize primary sources (original research, official statistics, company filings) over secondary (commentary, summaries)
   • Target a minimum of 5 distinct, authoritative sources per major claim
   • Record full citation details for every source: title, author/organization, publication date, URL, access date

4. Evaluate source quality:

   • Assess credibility: Is the author/organization authoritative in this domain?
   • Check recency: Is the data current enough for the research question?
   • Flag bias: Does the source have a commercial, political, or ideological interest in the claim?
   • Prefer: peer-reviewed journals, government statistics, established news outlets, recognized industry analysts
   • Flag: anonymous sources, self-published content without credentials, content from parties with financial stakes

5. Extract and organize findings:

   • For each sub-question: note the key finding, supporting evidence, and source
   • Identify where sources agree (consensus), where they conflict (controversy), and where data is absent (gap)
   • Flag hedging language in sources ("suggests", "may", "limited evidence") — do not upgrade their certainty

6. Synthesize across sources:

   • Group findings by theme, not by source
   • Highlight the weight of evidence: how many sources support each claim, and how strong is the evidence?
   • Note meaningful disagreements between sources and, where possible, explain why they diverge

7. Produce the research report:

   • Structure: Executive Summary → Background → Key Findings (by sub-question) → Synthesis → Gaps & Limitations → References
   • Inline citations for every factual claim: [Author, Year] or numbered footnotes
   • Confidence level per major claim: High / Medium / Low based on source quality and consensus

Output Format

## Research Report: [Topic]
**Date:** June 1, 2025 | **Depth:** Overview | **Sources reviewed:** 14

### Executive Summary
[3–5 sentence overview of the most important findings]

### Background
[Brief context for why this topic matters and how the research question was framed]

### Key Findings

#### 1. [Sub-question / theme]
> **Finding:** [Clear, declarative statement of the finding]
> **Evidence:** [Supporting data, statistics, or quotes from sources]
> **Sources:** [Smith et al., 2024]; [McKinsey, 2023]
> **Confidence:** High / Medium / Low

#### 2. [Sub-question / theme]
...

### Synthesis
[Overarching narrative connecting the findings; where consensus exists; where experts disagree]

### Gaps & Limitations
- [Key question the research could not answer due to lack of sources]
- [Notable caveat or methodological limitation in the available data]

### References
1. Smith, J. & Lee, A. (2024). *Title of Paper*. Journal of X. https://doi.org/...
2. McKinsey & Company (2023). *Industry Report Title*. https://mckinsey.com/...

Examples

Example Input

Research the current state of AI adoption in healthcare. Focus on: how widely it's being used, what the main use cases are, and what the biggest barriers to adoption are. I need a concise brief for a board presentation.

Example Output

## Research Brief: AI Adoption in Healthcare (2025)

### Executive Summary
AI is increasingly embedded in clinical workflows, with global healthcare AI spending projected at $45B in 2025. Radiology, diagnostics, and administrative automation lead adoption. Key barriers are regulatory uncertainty, data interoperability issues, and clinician trust deficits.

### Key Findings

**1. Adoption Level**
62% of large health systems have at least one AI application in production [Accenture, 2024]. Adoption is concentrated in large academic medical centers; community hospitals lag by 3–5 years [NEJM Catalyst, 2024]. Confidence: High.

**2. Top Use Cases**
- Medical imaging analysis (radiology, pathology): FDA-cleared AI tools now exceed 700 [FDA, 2025]
- Clinical documentation automation: reduces physician note time by 35% on average [AMA, 2024]
- Predictive risk scoring (sepsis, readmission): deployed in 38% of large ICUs [JAMA, 2023]

**3. Barriers to Adoption**
- Regulatory: FDA clearance process is slow; EU AI Act adds compliance overhead [Brookings, 2024]
- Data: 70% of health systems cite EHR interoperability as the #1 technical barrier [CHIME, 2024]
- Trust: Only 38% of clinicians trust AI recommendations without explanation [Lancet Digital Health, 2024]

### References
1. Accenture (2024). *Digital Health Technology Vision 2024*. https://accenture.com/...
2. FDA (2025). *AI/ML-enabled Medical Devices*. https://fda.gov/...

Boundaries

• Do NOT fabricate sources, authors, statistics, or citations — only report information from sources actually retrieved.
• Always cite claims; do NOT present findings as established facts if sources are limited or conflicting.
• Flag the recency of data — a statistic from 2019 may not reflect the current state.
• Do NOT present the view of a single source as consensus — always cross-reference major claims.
• When a research question falls outside publicly available information, acknowledge the gap rather than speculating.
• For medical, legal, or financial research: explicitly note that findings do not constitute professional advice and recommend qualified expert review.