Research Ideation Generator

What This Does

This playbook helps generate research ideas systematically. Instead of staring at a blank page, use structured prompts to generate research questions, hypotheses, and empirical strategies. Claude pushes beyond obvious approaches to surface non-obvious ideas.

Prerequisites

• Claude Code installed and configured
• A research area or topic to explore

The CLAUDE.md Template

Copy this into a CLAUDE.md file in your research folder:

# Research Ideation System

## Command
`/research-ideation [topic]` — Generate research ideas for a topic

## Ideation Framework

### Phase 1: Problem Space Exploration
Before generating ideas, understand the space:
1. What is the core phenomenon?
2. Why does it matter?
3. What's the current state of knowledge?
4. What are the key debates/tensions?

### Phase 2: Research Question Generation
Generate questions across dimensions:

**Descriptive Questions** (What is?)
- What is the prevalence/distribution of X?
- How does X vary across contexts?
- What are the components/dimensions of X?

**Causal Questions** (What causes?)
- Does X cause Y?
- What mediates the X→Y relationship?
- What moderates the X→Y relationship?

**Mechanism Questions** (How?)
- How does X produce Y?
- What is the process by which X operates?
- Why does X work in some contexts but not others?

**Normative Questions** (What should?)
- What is the optimal level of X?
- How should we design X?
- What interventions would improve X?

### Phase 3: Hypothesis Generation
For each research question, generate:
1. **Conventional hypothesis**: What most people would predict
2. **Contrarian hypothesis**: Opposite of conventional
3. **Contingent hypothesis**: "It depends on Z"
4. **Novel hypothesis**: Non-obvious prediction

### Phase 4: Empirical Strategy Brainstorm
For promising hypotheses:
- What data would test this?
- What's the ideal research design?
- What's a feasible alternative design?
- What are the main identification threats?

## Ideation Techniques

### Inversion
What if the opposite of conventional wisdom is true?

### Analogy Transfer
What works in field Y that hasn't been applied to field X?

### Boundary Exploration
What happens at the extremes? What's the smallest unit of analysis?

### Mechanism Deep Dive
Pick any relationship and ask "but how, exactly?"

### Counterfactual Thinking
What would the world look like if X didn't exist?

### Combination
What happens when A and B interact?

## Output Format

Research Ideation: [Topic]

Problem Space

[Brief characterization]

Research Questions

1. [Question 1] — [Type: Descriptive/Causal/Mechanism/Normative]
2. [Question 2] — [Type]
3. [Question 3] — [Type] ...

Most Promising Hypotheses

Hypothesis 1: [Statement]

• Rationale: [Why this might be true]
• Test: [How to test it]

Hypothesis 2: [Statement]

• Rationale: [Why]
• Test: [How]

Non-Obvious Ideas

• [Idea that isn't immediately obvious]
• [Contrarian take]
• [Cross-domain insight]

Gaps Identified

• [What hasn't been studied]
• [What's understudied]

## Quality Checks

Good research ideas should be:
- [ ] **Interesting**: Would people care about the answer?
- [ ] **Novel**: Not already answered definitively
- [ ] **Testable**: Can be empirically investigated
- [ ] **Meaningful**: Results would change how we think or act

Step-by-Step Setup

Step 1: Add to your research CLAUDE.md

Add the template above to your research folder's CLAUDE.md.

Step 2: Test with a topic

/research-ideation [your research area]

Or simply:

Generate research ideas about [topic]. Use the ideation framework.

Example Usage

User:

"/research-ideation AI adoption in healthcare"

Claude:

Research Ideation: AI Adoption in Healthcare

Problem Space

AI tools (diagnostic, administrative, clinical decision support) are becoming available in healthcare, but adoption is uneven. Key tensions: accuracy vs explainability, efficiency vs job displacement, innovation vs regulation.

Research Questions

Descriptive

1. What predicts which hospitals adopt AI early vs late? — Descriptive
2. How do clinicians' mental models of AI accuracy compare to actual accuracy? — Descriptive

Causal 3. Does AI diagnostic assistance improve or harm physician decision-making? — Causal 4. Does exposure to AI errors affect future trust calibration? — Causal

Mechanism 5. How do clinicians integrate AI recommendations with their own judgment? — Mechanism 6. Why do some physicians override AI more than others? — Mechanism

Normative 7. What level of AI explanation is optimal for different clinical decisions? — Normative 8. How should liability be allocated when AI contributes to errors? — Normative

Most Promising Hypotheses

Hypothesis 1: AI assistance helps average physicians more than experts

• Rationale: Experts have strong priors; AI provides less marginal information
• Test: Compare diagnostic accuracy improvement across experience levels

Hypothesis 2: AI explanation length has an inverse-U relationship with trust

• Rationale: Too little = black box distrust; too much = cognitive overload
• Test: Experiment varying explanation detail, measuring adoption

Hypothesis 3: AI adoption is higher for "defensive" diagnoses (ruling out) than "offensive" (ruling in)

• Rationale: Liability asymmetry — missing a diagnosis is worse than over-testing
• Test: Compare adoption rates by diagnosis type

Non-Obvious Ideas

• Automation complacency as hidden cost: Physicians may stop developing skills they delegate to AI, creating fragility
• AI as scapegoat: Physicians might adopt AI not for accuracy but for blame-shifting ("the AI suggested it")
• Inverse selection: Hospitals that most need AI (under-resourced, overworked) may be least able to implement it

Gaps Identified

• Long-term effects on physician skill development
• Patient preferences for AI involvement in their care
• How AI errors affect downstream patient trust in medicine broadly

Tips

• Push for contrarian ideas: Explicitly ask "What if the opposite were true?"
• Cross-domain insights: Ask "What does [other field] know about this that healthcare hasn't applied?"
• Quantity first: Generate 10-15 questions before evaluating. Don't self-censor early.
• Test feasibility later: Ideation phase is about possibility, not practicality.

Troubleshooting

Problem: Ideas feel generic/obvious

Solution: Go deeper. For any idea, ask "Yes, but specifically how?" or "What's the second-order effect?"

Problem: Can't generate contrarian ideas

Solution: List the assumptions behind conventional wisdom. What if each assumption were wrong?

Problem: Ideas aren't testable

Solution: Add "How would you test this?" as a required component for each hypothesis. Untestable ideas get cut.