Foundations
Prompting for Analysis and Research
Prompting for Analysis and Research
Analysis and research are core knowledge work tasks. You need to take data or information and extract meaning: What’s happening? Why does it matter? What should we do about it? LLMs excel at these tasks, but they require careful prompting. A vague analysis request produces generic insights. A structured analytical prompt produces actionable findings. You’re going to learn how to prompt the model to think like a researcher or analyst—methodical, evidence-based, and insightful.
Structuring Analytical Prompts: Data → Analysis → Insights → Recommendations
The best analytical prompts follow a clear flow: here’s the data, here’s what to analyze, here’s how to structure findings, here’s what decisions to inform.
The Four-Step Analytical Framework
STEP 1: Data Input
- What information are we analyzing?
- Context: Where did it come from? What's the time frame?
STEP 2: Analysis Focus
- What specific questions should we answer?
- What dimensions matter most?
STEP 3: Insights
- What patterns or findings emerge from the data?
- What's surprising or noteworthy?
STEP 4: Recommendations
- Given these insights, what should we do?
- What decisions do these findings inform?Example: Analyzing Sales Data
CONTEXT:
We've collected sales data for Q3. We need to understand:
- Performance against targets
- Customer trends
- Areas for improvement
DATA:
[Sales figures, customer segments, regional breakdown, etc.]
ANALYSIS FOCUS:
1. Which regions exceeded targets? Which underperformed?
2. How did different customer segments perform?
3. What products drove growth? What declined?
4. Are there patterns in customer acquisition vs. retention?
STRUCTURE YOUR ANSWER:
1. Executive summary (key findings in 3-4 bullets)
2. Regional analysis (by region, vs. target, trend)
3. Segment analysis (which types of customers drove growth)
4. Product analysis (growth areas and problem areas)
5. Recommendations (what to focus on in Q4)
FORMAT:
Use specific numbers. Avoid vague language like "improved" or "grew".
Be specific: "Revenue increased 23% YoY" not "Revenue grew significantly"Summarization Techniques: Extractive vs. Abstractive
When you need to condense information, there are two approaches.
Extractive Summarization
Pull the most important passages directly from the source.
Best for: Identifying key points without changing language Use when: You want to preserve exact wording, ensure accuracy
Prompt:
TASK: Extract the 5 most important sentences from this text.
These sentences should:
- Convey the main idea
- Be self-contained (understandable without context)
- Appear in order from the original
TEXT:
[Paste the full text]
RETURN: 5 sentences only, numbered, in original orderAbstractive Summarization
Rewrite the content in your own words, condensing it.
Best for: Creating readable summaries that flow naturally Use when: You want conciseness and clarity over exact wording
Prompt:
TASK: Summarize this document in your own words.
Requirements:
- 200 words (approximate)
- Cover: Main argument, key evidence, conclusion
- Tone: Accessible to a [audience type]
- Flow: As a cohesive summary, not a list
AVOID:
- Direct quotes (paraphrase instead)
- Jargon without explanation
- Side topics or tangents
- "In summary..." phrases
DOCUMENT:
[Paste content]Hybrid: Extractive Headlines + Abstractive Summary
Often you want both: key quotes plus a condensed overview.
TASK: Summarize this article with both structure and detail.
FORMAT:
1. Main Headline: [1 sentence capturing the core idea]
2. Key Quote: [1 meaningful direct quote from the article]
3. What Happened: [2-3 sentences explaining the situation]
4. Why It Matters: [1-2 sentences on implications]
5. What's Next: [1-2 sentences on expected developments]
ARTICLE:
[Paste content]Comparative Analysis: Showing Relationships and Trade-Offs
Comparing options is a frequent analytical task. Structure these prompts carefully.
Side-by-Side Comparison
Use a table or list structure to show how options differ.
TASK: Compare [Option A] vs [Option B]
DIMENSIONS TO COMPARE:
- Cost (upfront and recurring)
- Ease of use/learning curve
- Scalability (as we grow)
- Integration with existing tools
- Support quality
- Community/ecosystem
- [Add domain-specific dimensions]
FORMAT: Table with:
- Rows: Comparison dimensions
- Columns: Each option
- Cells: Specific facts or assessments
Add a summary: Which is better for which use case?
INSTRUCTIONS:
- Be specific (not "expensive" but "$X/month")
- Be fair (acknowledge strengths of both)
- Include trade-offs (can't be great at everything)Trade-off Analysis
Highlight what you gain and lose with each option.
TASK: Analyze the trade-offs between:
- Option A: [Brief description]
- Option B: [Brief description]
For each option, identify:
1. Key advantages (2-3 main benefits)
2. Key disadvantages (2-3 main drawbacks)
3. Best scenario (when this is the right choice)
4. Worst scenario (when this would be problematic)
FORMAT: Clear comparison showing:
- What you gain vs. what you lose
- Quantified impact where possible
- What decision this favors
CONTEXT:
[Explain decision constraints: budget, timeline, team skills, etc.]
RECOMMENDATION:
Based on our priorities, which is better? Why?Research Synthesis: Combining Multiple Sources
When you have multiple sources of information, synthesis combines them into a coherent picture.
Research Synthesis Pattern
TASK: Synthesize findings from multiple sources on [topic]
SOURCES:
Source 1: [content or summary]
Source 2: [content or summary]
Source 3: [content or summary]
SYNTHESIS QUESTIONS:
1. What do all sources agree on?
2. Where do sources disagree or offer different perspectives?
3. What do they emphasize differently?
4. What's the most complete picture from combining them?
OUTPUT:
1. Points of agreement (consensus findings)
2. Points of disagreement (conflicting views)
3. Integrated summary (combining the perspectives)
4. Confidence level (how confident in these findings?)
5. What remains unclear (gaps in available information)Real-World Example: Synthesizing Expert Opinions
I've collected advice on launching a startup from three sources.
SOURCE 1 - Paul Graham (YCombinator):
[Key quotes or summary of his advice]
SOURCE 2 - Reid Hoffman (LinkedIn founder):
[Key quotes or summary of his advice]
SOURCE 3 - Sheryl Sandberg (Meta COO):
[Key quotes or summary of her advice]
SYNTHESIS TASK:
1. What principles do all three emphasize?
2. Where do they diverge in advice?
3. Create an integrated guide: If I follow the best of all three, what do I do?
4. Are there contradictions? How would you resolve them?
RETURN: Synthesized startup advice combining all three perspectivesFact-Checking and Verification Prompts
When accuracy matters, build verification into your prompt.
Fact-Checking Prompt
TASK: Check the factual accuracy of these statements.
INSTRUCTIONS:
- For each statement, note if you're confident it's true, false, or uncertain
- Explain your reasoning
- If uncertain, note what sources would confirm this
- Flag any claims that depend on current information (dates, statistics)
STATEMENTS TO CHECK:
1. [Statement 1]
2. [Statement 2]
3. [Statement 3]
FORMAT:
Statement | Assessment | Reasoning | Confidence Level
CRITICAL NOTE:
I'm asking you to note where you're uncertain. If you're not sure about
something, say so. This is more valuable than a guess.Verification Against Source
Use this when you have a primary source to check against.
TASK: Verify this claim against the provided source document.
CLAIM:
[The statement being checked]
SOURCE DOCUMENT:
[The authoritative source text]
VERIFICATION:
1. Does the source support this claim?
2. Is it a direct quote, paraphrase, or inference?
3. Are there qualifications in the source that change the meaning?
4. If the claim is true, what specific passage supports it?
ASSESSMENT:
- Supported (with evidence)
- Partially supported (with caveats)
- Contradicted (source says something different)
- Not addressed (source doesn't cover this)Building Multi-Step Research Workflows
Complex research often requires multiple steps: gather, organize, analyze, synthesize, conclude.
Research Workflow Example: Competitive Analysis
class CompetitiveAnalyzer:
"""Conduct multi-step competitive analysis"""
def __init__(self):
self.client = anthropic.Anthropic()
def step_1_gather_competitor_info(self, competitors_list):
"""Collect key information on each competitor"""
response = self.client.messages.create(
model="claude-3-5-sonnet-20241022",
max_tokens=1500,
messages=[{
"role": "user",
"content": f"""Provide an overview of these competitors:
{competitors_list}
For each, briefly note:
- Core product/offering
- Target market
- Pricing model (if known)
- Key strengths
- Known weaknesses
Format as a quick reference table."""
}]
)
return response.content[0].text
def step_2_identify_gaps(self, competitor_info, our_product):
"""Find gaps between competitors and us"""
response = self.client.messages.create(
model="claude-3-5-sonnet-20241022",
max_tokens=1000,
messages=[{
"role": "user",
"content": f"""Based on this competitive landscape:
{competitor_info}
And our product: {our_product}
Identify:
1. Market gaps (unmet needs)
2. Feature gaps (what competitors don't offer)
3. Positioning gaps (how we can differentiate)
4. Customer segment gaps (who competitors don't serve)
Format as clear list."""
}]
)
return response.content[0].text
def step_3_analyze_strengths(self, competitor_info, our_strengths):
"""What are our competitive advantages?"""
response = self.client.messages.create(
model="claude-3-5-sonnet-20241022",
max_tokens=800,
messages=[{
"role": "user",
"content": f"""Given this competitive landscape:
{competitor_info}
And our key strengths: {our_strengths}
What are our genuine competitive advantages?
What can we uniquely offer that competitors can't easily copy?
Where should we focus to win?
Be specific. Avoid generic claims."""
}]
)
return response.content[0].text
def step_4_synthesize_strategy(self, gaps, advantages):
"""Create strategic recommendations"""
response = self.client.messages.create(
model="claude-3-5-sonnet-20241022",
max_tokens=1200,
messages=[{
"role": "user",
"content": f"""Based on this competitive analysis:
MARKET GAPS:
{gaps}
OUR ADVANTAGES:
{advantages}
Create a go-to-market strategy that:
1. Targets the gaps where we're strong
2. Leverages our advantages
3. Avoids head-to-head competition in their strengths
4. Positions us uniquely
Include:
- Target customer segment (specific)
- Key messaging (how we're different)
- Feature prioritization (what to build/emphasize)
- Potential partnerships or tactics"""
}]
)
return response.content[0].text
def run_analysis(self, competitors, our_product, our_strengths):
"""Run the complete competitive analysis"""
print("Step 1: Gathering competitor information...")
competitor_info = self.step_1_gather_competitor_info(competitors)
print("✓ Complete\n")
print("Step 2: Identifying market gaps...")
gaps = self.step_2_identify_gaps(competitor_info, our_product)
print("✓ Complete\n")
print("Step 3: Analyzing our strengths...")
advantages = self.step_3_analyze_strengths(competitor_info, our_strengths)
print("✓ Complete\n")
print("Step 4: Creating strategy...")
strategy = self.step_4_synthesize_strategy(gaps, advantages)
print("✓ Complete\n")
return {
"competitors": competitor_info,
"gaps": gaps,
"advantages": advantages,
"strategy": strategy
}
# Usage
analyzer = CompetitiveAnalyzer()
analysis = analyzer.run_analysis(
competitors="Competitor A, Competitor B, Competitor C",
our_product="Our AI-powered analytics platform",
our_strengths="Unique real-time processing, proprietary ML models, enterprise support"
)Handling Uncertainty in Analysis
Analytical prompts should acknowledge uncertainty rather than pretend to know things.
Uncertainty-Aware Analysis Prompt
TASK: Analyze this situation and identify what you're confident about vs.
what you're uncertain about.
SITUATION:
[Description of what needs analysis]
DATA PROVIDED:
[What data or information do you have?]
ANALYSIS:
1. What can you say with confidence? (Based on clear evidence)
2. What are you uncertain about? (Missing data or conflicting info)
3. For uncertain areas, what additional information would help?
FORMAT:
- High confidence findings (strong evidence)
- Medium confidence assessments (reasonable inference)
- Low confidence or speculative (limited evidence)
CRITICAL: Make uncertainty explicit. This is more valuable than false confidence.Key Takeaway
Analytical prompts should structure data → analysis → insights → recommendations. Use extractive summarization to pull key points, abstractive to condense naturally. Comparative analysis requires clear dimensions and trade-off visibility. Synthesize multiple sources by noting agreement, disagreement, and integrated conclusions. Build verification into important claims. Multi-step research workflows break complex analysis into manageable steps.
Exercise: Conduct a Case Study Analysis Using Structured Prompts
Your task is to analyze a real or fictional business case using the analytical framework.
The Scenario
A mid-size SaaS company is facing a challenge:
Company: CloudSync (project management software) Problem: Revenue growth has slowed from 40% YoY to 18% YoY Data Available: [You’ll create/imagine realistic metrics]
Your Task
Design a 4-step analytical prompt sequence that:
-
Step 1: Diagnose the Problem
- What changed? (growth metrics, customer metrics, market conditions)
- What patterns emerge?
-
Step 2: Compare to Alternatives
- How do we compare to competitors on growth?
- Are there industry trends we’re missing?
-
Step 3: Synthesize Insights
- What are the root causes?
- What’s in our control vs. external?
-
Step 4: Recommend Actions
- What should the company do?
- Prioritize by impact and feasibility
Deliverable
For each step, provide:
- The Prompt: Full text of what you’d send to the model
- Expected Output: What kind of answer would be valuable?
- How It Builds on Previous Steps: How does this step use prior findings?
Example Start
## Step 1: Diagnose the Problem
**Goal:** Understand what's changed and why growth slowed
**The Prompt:**
CloudSync is a project management SaaS company. Revenue growth has slowed from
40% YoY to 18% YoY over the past year.
Data:
- CAC (customer acquisition cost): $500 (unchanged)
- LTV (lifetime value): $12,000 (down 15%)
- Churn rate: 5% (up from 2%)
- NPS: 32 (down from 58)
[More data...]
Analyze:
1. What metrics changed most significantly?
2. What patterns suggest root causes?
3. Which are symptoms vs. root causes?
Format: Clear analysis with specifics (numbers and trends)
**Expected Output:**
- Specific metric changes and their significance
- Root cause hypotheses (multiple)
- Early warning indicators we might have missedBonus Challenge
After designing your 4-step sequence:
- Imagine the output of Step 1
- Design a follow-up “Step 2B” that dives deeper into the most surprising finding
- Show how iteration and follow-up prompts reveal deeper insights
This demonstrates that analysis isn’t one-and-done; it’s iterative discovery.