Advanced AI Prompt Engineering Guide - Google-Style Methods
Uploaded by
Sabarna BiswasAdvanced AI Prompt Engineering Guide - Google-Style Methods
Uploaded by
Sabarna Biswas12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
🧠 Advanced AI Prompt Engineering Guide:
Google-Style Methods
1. CHAIN-OF-THOUGHT (CoT) PROMPTS
Best for: Complex reasoning, math problems, logical tasks
Basic CoT Template:
You are a senior mathematician. Solve this problem step by step:
PROBLEM: [Insert problem]
First, identify what's being asked.
Second, list relevant formulas/concepts.
Third, apply step-by-step logic.
Fourth, verify your answer.
Fifth, provide final answer clearly.
Advanced CoT with Self-Correction:
Act as a mathematics professor solving [topic] problems. Follow this framework:
1. Problem Interpretation: Restate the problem in your own words
2. Known Variables: List all given information
3. Required Formulas: Identify relevant equations
4. Step-by-Step Solution: Show calculations with units
5. Sanity Check: Does the answer make logical sense?
6. Alternative Method: Solve using different approach
7. Common Mistakes: List 2-3 errors students typically make
8. Final Boxed Answer: Present clearly
Problem: [Insert here]
[Link] 1/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
2. FEW-SHOT LEARNING PROMPTS
Best for: Pattern recognition, classification, consistent formatting
Multi-Example Template:
I'll show you examples of how to format answers, then you apply the same pattern
EXAMPLE 1:
Question: "What is 15% of 200?"
Solution: "15% means 15/100 = 0.15. Multiply: 0.15 × 200 = 30. Answer: 30"
EXAMPLE 2:
Question: "A shirt costs $40 with 20% discount. Final price?"
Solution: "Discount = 20% of 40 = 0.20 × 40 = $8. Final price = 40 - 8 = $32. An
Now solve:
Question: "Find 35% of 280"
[Follow the exact format from examples]
Role-Based Few-Shot:
You are an expert data analyst. Format responses exactly like this:
Example Analysis:
Dataset: Sales figures
Question: "What was Q3 growth?"
Response:
ANALYSIS
======
Data Points: Q2=$150K, Q3=$180K
Calculation: ((180-150)/150)×100 = 20%
Conclusion: Q3 growth was 20%
Now analyze:
Dataset: User engagement metrics
Question: "Calculate month-over-month increase"
[Follow format exactly]
[Link] 2/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
3. SYSTEM ROLE + CONSTRAINTS PROMPTS
Best for: Controlling tone, depth, format
Multi-Constraint Template:
SYSTEM PROMPT:
You are a Google DeepMind research assistant specializing in [topic]. Adhere to:
1. RESPONSE STRUCTURE:
- Start with "SUMMARY:" (1 sentence)
- Then "KEY POINTS:" (bullet points, max 5)
- Then "DETAILED ANALYSIS:" (paragraph)
- End with "REFERENCES:" (if applicable)
2. STYLE GUIDELINES:
- Use British English spelling
- Academic tone but accessible
- No markdown, plain text only
- Word limit: 250 words
3. CONTENT RULES:
- Cite recent studies (post-2020)
- Mention limitations
- Include practical applications
USER QUERY: [Insert here]
4. REACT (Reasoning + Acting) PROMPTS
Best for: Tool use, research, multi-step tasks
You are an AI research assistant with access to these tools:
1. CALCULATOR (for math)
2. SEARCH (for facts)
3. CODE_INTERPRETER (for data)
4. VALIDATOR (for checking)
TASK: Analyze the economic impact of AI automation
Follow this thought process:
[Link] 3/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
Thought: I need to break this into components
Action: Use SEARCH for "AI automation GDP studies 2023"
Observation: [AI fills this]
Thought: Now I need recent statistics
Action: Use SEARCH for "McKinsey AI productivity report 2024"
Observation: [AI fills this]
... continue until complete answer
5. PAL (Program-Aided Language) PROMPTS
Best for: Mathematical/computational reasoning
You are a Python-assisted mathematician. For any quantitative problem:
1. First, write Python code to solve
2. Execute code mentally (describe steps)
3. Show output
4. Interpret results in plain English
Problem: "If a car accelerates from 0 to 60 mph in 5 seconds, what's its acceler
Your response format:
STEP 1: UNDERSTAND
Convert mph to m/s: 60 mph = 26.8224 m/s
STEP 2: CODE
initial_velocity = 0
final_velocity = 26.8224 # m/s
time = 5 # seconds
acceleration = (final_velocity - initial_velocity) / time
print(f"Acceleration: {acceleration} m/s²")
STEP 3: OUTPUT
Acceleration: 5.36448 m/s²
STEP 4: ANSWER
The car accelerates at approximately 5.36 m/s².
[Link] 4/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
6. METICULOUS SCAFFOLDING PROMPT
Best for: Deep analysis, research papers, comprehensive answers
ANALYSIS REQUEST FORM
=====================
Topic: [User inserts topic]
INSTRUCTIONS: Process through these 7 layers:
LAYER 1: DECONSTRUCTION
- What are the explicit components?
- What are the implicit assumptions?
LAYER 2: CONTEXTUALIZATION
- Historical context
- Current state (2024)
- Future projections
LAYER 3: MULTIPERSPECTIVE
- Economic viewpoint
- Sociological angle
- Technological perspective
- Ethical considerations
LAYER 4: EVIDENCE GATHERING
- Statistical data needed
- Research papers to cite
- Expert opinions required
LAYER 5: SYNTHESIS
- Connect Layer 1-4 findings
- Identify patterns/contradictions
LAYER 6: APPLICATION
- Practical use cases
- Implementation challenges
- Risk assessment
LAYER 7: COMMUNICATION
- Executive summary (1 paragraph)
- Detailed report
- Visual representation suggestion
[Link] 5/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
BEGIN ANALYSIS NOW:
7. SELF-CRITIQUE PROMPT
Best for: Reducing hallucinations, improving accuracy
You are a Google fact-checking AI. For any response you generate:
1. Generate your initial answer
2. Then apply CRITIQUE PHASE:
- Identify 3 potential weaknesses in your answer
- List assumptions you made
- Note any missing data
- Flag uncertain statements
3. REVISION PHASE:
- Correct weaknesses
- Add caveats
- Improve precision
4. FINAL OUTPUT with this header:
✅ VERIFIED RESPONSE
Confidence: High/Medium/Low
Last Updated: [Assume current date]
Caveats: [List limitations]
[Final answer here]
Question: [Insert here]
8. CREATIVE CONSTRAINTS PROMPT
Best for: Writing, design, brainstorming
You are a creative director at Google Creative Lab.
TASK: [Creative task]
[Link] 6/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
CONSTRAINTS (use all):
1. Include exactly 3 metaphors
2. Use words from this list: [list]
3. Structure as dialogue between two historical figures
4. Include a sudden twist at 80% point
5. End with a question that challenges assumptions
ADDITIONAL RULES:
- Avoid clichés about [topic]
- Include one technical detail made accessible
- Emotion should shift from curiosity to insight
BEGIN:
PROMPT ENGINEERING BEST PRACTICES (Google Style)
A. The 5C Framework:
1. Clear — Unambiguous instructions
2. Concise — Remove redundancy
3. Contextual — Provide relevant background
4. Constrained — Set boundaries
5. Chained — Logical flow between steps
B. Testing Prompts:
Variant A: [Prompt style 1]
Variant B: [Prompt style 2]
Variant C: [Prompt style 3]
Evaluate which produces:
1. Most accurate answer
2. Best readability
3. Appropriate depth
4. Factual consistency
[Link] 7/8
12/23/25, 1:13 PM Advanced AI Prompt Engineering Guide: Google-Style Methods
C. Meta-Prompt for Improving Prompts:
Analyze this prompt: [Insert your prompt]
Rate on scale 1-10:
- Specificity: _
- Constraint clarity: _
- Role definition: _
- Output format: _
Suggest 3 improvements:
1. [Suggestion]
2. [Suggestion]
3. [Suggestion]
Rewrite the prompt incorporating improvements.
GOLDEN RULE FOR ENTERPRISE PROMPTS:
“Write prompts that generate prompts” — Create meta-prompts that help users build their
own effective prompts for specific domains.
Would you like domain-specific prompts (medical, legal, coding, etc.) using these methods?
[Link] 8/8