AI-Assisted Research

The Evolution of Desk Review

Traditionally, desk review meant hours of Google searches, scanning reports, and manually organizing findings. AI tools like ChatGPT and Claude have transformed this process—but only if used strategically.

AI as Assistant, Not Replacement

AI accelerates research and helps identify sources you might miss. But it doesn't replace critical thinking, source verification, or—most importantly—community validation. Think of AI as a research assistant who does the initial legwork, not the final decision-maker.

What is MCP (Model Context Protocol)?

MCP is a structured way to instruct AI assistants like ChatGPT or Claude, ensuring you get focused, useful outputs instead of generic responses.

Think of it as briefing a human research assistant. You wouldn't just say "Tell me about youth unemployment." You'd provide:

Context about your project and what you already know
Specific objectives for the research
Clear instructions about format and depth
Expectations around sources and evidence

MCP formalizes this approach into four structured blocks that you'll learn in detail in the next section.

Why Use AI for Problem Analysis?

When used correctly, AI-assisted research provides five key advantages:

⚡

Speed

Accelerate initial literature review and source identification from days to hours. Get a preliminary Problem Tree structure in 30-40 minutes instead of a full day.

🌐

Scope

Quickly survey a wide range of perspectives and evidence sources—UN reports, academic research, NGO publications—that you might not find through traditional search.

🗂️

Structure

Get organized outputs that fit your analytical framework. Request outputs in Problem Tree format, tables of indicators, or lists of validation questions—pre-formatted and ready to use.

📚

Sources

Identify credible reports and data you might have missed. AI can surface publications from international organizations, research institutions, and government agencies that don't rank highly in Google.

❓

Questions

Generate stakeholder validation questions systematically. For each assumption in your Problem Tree, AI can suggest 2-3 open-ended questions that guide meaningful community conversations.

What AI Can and Cannot Do

✅ AI Excels At:

Summarizing published research and identifying patterns across sources
Suggesting cause-effect relationships documented in literature
Formatting outputs into Problem Tree structure, tables, or lists
Generating initial hypotheses that need field validation
Drafting validation questions from your assumptions

❌ AI Cannot:

Replace community voices—it doesn't know your specific context lived experience
Verify current accuracy—it may cite outdated or generalized information
Understand cultural nuance—context-specific dynamics require local knowledge
Make ethical decisions—you must decide which causes to address and how
Replace your critical thinking—every output needs your review and validation

The 70/30 Rule

Use AI for 70% of the initial desk review work (literature scan, source identification, preliminary structure). Reserve 30% of your effort for verification, contextualization, and preparing for stakeholder validation. This balance maintains rigor while gaining efficiency.

The Research Workflow

Here's how AI-assisted research fits into your Problem Tree development:

Define your problem scope (15 minutes)—You decide what to investigate
Customize MCP prompt (10 minutes)—Provide context and objectives
Execute AI research (5 minutes)—Run the prompt, save outputs
Quality verification (20 minutes)—Check sources, flag assumptions
Build Problem Tree (25 minutes)—Organize findings, tag evidence
Prepare validation questions (15 minutes)—Convert assumptions to stakeholder questions

Total time: ~90 minutes for a preliminary Problem Tree ready for stakeholder validation.

Critical Success Factors

1. Be Specific in Your Prompts

Generic input = generic output. The more context you provide about your geography, population, and what you already know, the more useful the AI response.

2. Always Verify Sources

AI sometimes generates plausible-sounding but incorrect citations. Spot-check 3-5 key sources by opening the links or searching for the publications.

3. Tag Everything as (E) or (A)

Even if AI provides a claim, tag it (A) for assumption unless you've verified the source. Transparency builds credibility.

4. Prepare to Be Surprised

The goal of stakeholder validation isn't to confirm what AI told you—it's to learn what AI missed, got wrong, or oversimplified. Go to the community ready to be surprised.

Next Step: Deep Dive into MCP

Now that you understand why and when to use AI-assisted research, you're ready to learn the Model Context Protocol structure in detail. The next page provides a complete template you can copy, customize, and use immediately.

Ethical Considerations

Data Privacy

Don't include personally identifiable information about community members in your prompts. Keep prompts general and focused on published research.

Bias Awareness

AI models reflect biases in their training data. Be especially critical when researching issues related to marginalized communities, where historical bias in published literature is common.

Community Partnership

Always position AI research as preliminary and stakeholder engagement as definitive. Make it clear in your proposals and conversations that community voices shaped your final analysis.

Ready for the Deep Dive?

You now understand the "why" and "what" of AI-assisted research. Move to Model Context Protocol to get the complete four-block template and learn exactly how to structure your prompts for maximum effectiveness.