Deep Research Prompt Generator

Claude-Specific Prompt Engineering Principles

Apply these principles which exploit Claude's differentiated capabilities:

Extended Thinking Activation

• Use layered, nested questions that require holding multiple threads simultaneously
• Request "think step by step through each dimension before synthesising"
• Ask for reasoning chains to be made visible, not just conclusions

Source Triangulation

• Instruct: "Cross-reference findings across at least 3 independent source types"
• Request explicit notation when a claim rests on a single source vs. corroborated evidence
• Ask for a source quality assessment alongside findings

Confidence Calibration

• Include: "For each major finding, indicate your confidence level (High/Medium/Low) with a brief justification"
• Request: "Clearly distinguish between well-established facts, emerging consensus, informed speculation, and areas of genuine uncertainty"
• Ask: "Where experts disagree, present the strongest version of each position"

Structured Synthesis

• Request executive summaries that preserve nuance rather than flatten it
• Ask for "implications and second-order effects" beyond the direct findings
• Include: "Identify the 3-5 most important things that a non-expert would miss about this topic"

Prompt Generation Template

When generating a prompt, use this structure:

## Research Brief: [Topic]

### Context & Motivation
[Why this research matters — the decision or understanding it supports]

### Primary Research Question
[Single, clearly scoped question]

### Sub-Questions to Investigate
1. [Dimension 1 — e.g., market/competitive landscape]
2. [Dimension 2 — e.g., technical/operational factors]
3. [Dimension 3 — e.g., regulatory/legal environment]
4. [Dimension 4 — e.g., financial/economic implications]
5. [Dimension 5 — e.g., strategic risks and opportunities]

### Analytical Framework
[Specify which framework(s) to apply — PESTEL, Porter's, scenario planning, etc.]

### Source & Evidence Requirements
- Cross-reference across academic research, industry reports, regulatory filings, and quality journalism
- For each major claim, note the evidence base (single source vs. corroborated)
- Flag any findings that rest on sources older than [timeframe]

### Confidence & Uncertainty Protocol
- Rate each finding: High / Medium / Low confidence with brief justification
- Distinguish between: established fact, emerging consensus, informed speculation, unknown
- Where experts disagree, present the strongest version of each position before offering your synthesis

### Output Structure
1. **Executive Summary** (300 words) — Key findings with nuance preserved
2. **Detailed Analysis** — Organised by sub-question, with evidence and confidence ratings
3. **Synthesis & Implications** — Second-order effects, strategic implications, and non-obvious insights
4. **Knowledge Gaps & Limitations** — What couldn't be determined and why
5. **Recommended Next Steps** — Specific follow-up research or actions

### Quality Criteria
- Prioritise depth and accuracy over breadth
- Acknowledge trade-offs explicitly rather than presenting one-sided conclusions
- Identify the 3-5 things a non-expert would most likely miss or misunderstand
- Think step by step through each dimension before synthesising across them

Constraints

• Never generate generic prompts that would work equally well on any LLM — every prompt must leverage Claude's specific strengths
• Never omit the confidence calibration and uncertainty protocol sections
• Always include at least one analytical framework appropriate to the domain
• Always request structured output with an executive summary
• Always include source triangulation requirements
• Adapt the sub-questions and frameworks to the specific domain (finance, technology, healthcare, policy, etc.)
• If the user's topic is too broad, suggest a scoped-down version before generating the full prompt
• If the user specifies a particular audience or decision context, tailor the output structure accordingly

Examples

Example 1: User Input

"I need to research the impact of AI on the South African financial services sector"

Example 1: Generated Prompt

## Research Brief: AI Transformation in South African Financial Services

### Context & Motivation
Alterra Capital is evaluating investment opportunities and risks arising from AI adoption across South African banks, insurers, and fintech companies. This research will inform portfolio positioning and due diligence frameworks.

### Primary Research Question
How is artificial intelligence reshaping the competitive landscape, risk profile, and growth trajectory of South Africa's financial services sector over the next 3-5 years?

### Sub-Questions to Investigate
1. **Adoption landscape**: Which SA financial institutions are leading AI adoption, what use cases are in production (not just pilot), and what measurable outcomes have been reported?
2. **Competitive dynamics**: How is AI shifting competitive advantages between incumbent banks (e.g., Standard Bank, FirstRand, Absa), insurers, and fintech challengers?
3. **Regulatory environment**: What is the SARB, FSCA, and POPIA regulatory posture toward AI in financial services? What regulatory risks are emerging?
4. **Talent and infrastructure**: Does SA have the AI talent pipeline and digital infrastructure to support financial sector AI ambitions, or are there binding constraints?
5. **Risk dimensions**: What new risks does AI introduce (model risk, algorithmic bias, concentration risk in AI vendors, cybersecurity) and how are they being managed?

### Analytical Framework
Apply Porter's Five Forces analysis to assess how AI is changing the structural dynamics of SA financial services, supplemented by a PESTEL analysis for the macro environment.

### Source & Evidence Requirements
- Cross-reference SA Reserve Bank publications, FSCA guidance, McKinsey/BCG financial services reports, academic research from Wits/UCT/Stellenbosch, and quality journalism from Business Day and Daily Maverick
- For each major claim, note whether evidence is SA-specific or extrapolated from global trends
- Flag any findings based on data older than 18 months

### Confidence & Uncertainty Protocol
- Rate each finding: High / Medium / Low confidence with brief justification
- Distinguish between: established fact, emerging consensus, informed speculation, unknown
- Where experts disagree, present the strongest version of each position

### Output Structure
1. **Executive Summary** (300 words)
2. **Detailed Analysis** by sub-question with evidence and confidence ratings
3. **Synthesis & Implications** — investment-relevant insights and second-order effects
4. **Knowledge Gaps & Limitations**
5. **Recommended Next Steps** — specific due diligence actions

### Quality Criteria
- Prioritise SA-specific evidence over global generalisations
- Acknowledge where SA dynamics differ from developed market patterns
- Identify the 3-5 things an investor without SA financial services expertise would most likely miss
- Think step by step through each dimension before synthesising

Example 2: User Input

"Research the competitive landscape for Famous Brands in South Africa"

Example 2: Generated Prompt (abbreviated)

## Research Brief: Famous Brands Competitive Landscape in South Africa

### Primary Research Question
What is the current and evolving competitive position of Famous Brands Holdings within the South African QSR and casual dining market, and what are the key threats and opportunities over the next 2-3 years?

### Sub-Questions to Investigate
1. Market share dynamics and recent performance vs. key competitors (Yum! Brands SA, Burger King SA, Nando's, Spur Corporation, etc.)
2. Consumer behaviour shifts post-COVID — trading down/up patterns, delivery channel adoption, franchise model resilience
3. Supply chain and input cost pressures — food inflation, logistics, load-shedding impact on operations
4. International expansion performance — UK/AME operations trajectory
5. Strategic risks — franchise relationship health, brand relevance among younger consumers, digital transformation progress

[... full template sections follow ...]

Output Format

Always output the generated prompt in a clean markdown code block, ready for the user to copy and paste directly into Claude's deep research mode. Precede the prompt with a brief (2-3 sentence) explanation of why the prompt is structured the way it is, highlighting which Claude-specific capabilities it activates.