Advanced Prompt Patterns

Prompt:
I need to use an AI to analyze customer support
tickets and categorize them by urgency, topic,
and sentiment. The AI should also suggest a
response template.

Design the optimal prompt for this task. Include:
- A clear role assignment
- Specific instructions with edge cases
- Output format specification
- 2-3 few-shot examples
- Error handling instructions

The AI will generate a complete, optimized prompt
that you can then use directly for your task.

Prompt:
Generate a response to the user's question below.
Then evaluate your response against these principles:

Principles:
1. Accuracy: All facts must be verifiable
2. Completeness: Address all parts of the question
3. Clarity: Use simple language, avoid jargon
4. Balance: Present multiple perspectives if relevant
5. Safety: Do not include harmful or misleading info

After generating your response, score it 1-5 on each
principle. If any score is below 4, revise the
response to improve that aspect. Show both the
original and revised response.

User Question: "What are the pros and cons of
nuclear energy?"

Prompt:
Extract the following information from this job
posting and return it as a JSON object:

Fields: title, company, location, salary_range,
required_skills (array), experience_years,
remote_policy (onsite/hybrid/remote)

Job Posting: """
Senior React Developer at TechCorp in Austin, TX.
$130K-$160K. Hybrid (3 days office). 5+ years
experience required. Must know React, TypeScript,
Node.js, PostgreSQL, and AWS.
"""

Response:
{
  "title": "Senior React Developer",
  "company": "TechCorp",
  "location": "Austin, TX",
  "salary_range": "$130K-$160K",
  "required_skills": ["React", "TypeScript",
    "Node.js", "PostgreSQL", "AWS"],
  "experience_years": 5,
  "remote_policy": "hybrid"
}

Prompt:
Analyze this text and return the results in XML
format with tags for: <summary>, <key_points>,
<sentiment>, and <action_items>.

Response:
<analysis>
  <summary>The Q3 report shows 15% revenue
  growth but rising customer churn.</summary>
  <key_points>
    <point>Revenue up 15% YoY</point>
    <point>Churn rate increased to 8%</point>
  </key_points>
  <sentiment>mixed</sentiment>
  <action_items>
    <item>Investigate churn causes</item>
    <item>Launch retention campaign</item>
  </action_items>
</analysis>

# YAML - great for configuration-like output
Prompt: "Generate a deployment config in YAML format"

Response:
app:
  name: my-service
  version: 2.1.0
  replicas: 3
  resources:
    cpu: "500m"
    memory: "256Mi"

# CSV - great for tabular data
Prompt: "List top 5 programming languages with their
use cases. Format as CSV with headers."

Response:
language,primary_use,popularity_rank
Python,Data Science/AI,1
JavaScript,Web Development,2
TypeScript,Full-Stack Apps,3
Java,Enterprise/Android,4
Rust,Systems Programming,5

def create_review_prompt(product, review_text, aspects):
    """Generate a product review analysis prompt."""
    return f"""
Analyze the following {product} review.

Review: "{review_text}"

For each of these aspects, provide a rating (1-5)
and a brief explanation:
{chr(10).join(f'- {a}' for a in aspects)}

Format your response as JSON with this structure:
{{
  "overall_sentiment": "positive/negative/neutral",
  "aspects": {{
    "aspect_name": {{
      "rating": 1-5,
      "explanation": "brief explanation"
    }}
  }},
  "summary": "one-sentence summary"
}}
"""

# Usage
prompt = create_review_prompt(
    product="laptop",
    review_text="Great performance but battery life
    is disappointing. The keyboard feels premium.",
    aspects=["performance", "battery", "build_quality"]
)

def build_dynamic_prompt(user):
    # Adjust complexity based on user expertise
    if user.level == "beginner":
        style = "Use simple language, avoid jargon,
        include analogies"
    elif user.level == "intermediate":
        style = "Use technical terms with brief
        explanations"
    else:
        style = "Use precise technical language,
        assume deep domain knowledge"

    # Add context from previous interactions
    history = get_relevant_history(user.id, limit=3)

    return f"""
{style}

Previous context from this user:
{history}

User's current question: {user.query}
"""

# Using Claude's API with an image
import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{
        "role": "user",
        "content": [
            {
                "type": "image",
                "source": {
                    "type": "url",
                    "url": "https://example.com/chart.png"
                }
            },
            {
                "type": "text",
                "text": "Analyze this sales chart.
                Identify: (1) overall trend, (2) seasonal
                patterns, (3) anomalies, (4) forecast
                for next quarter. Format as JSON."
            }
        ]
    }]
)

Prompt:
Answer the user's question using ONLY the provided
context documents. If the answer is not in the
documents, say "I don't have enough information to
answer this question."

Context Documents:
---
Document 1 (Company Policy - Updated Jan 2026):
[retrieved content here]
---
Document 2 (Employee Handbook Section 4.2):
[retrieved content here]
---

User Question: "What is the company's policy on
remote work for new employees?"

Instructions:
- Cite the specific document and section
- Quote relevant passages
- If documents conflict, note the discrepancy
- Do not use knowledge outside these documents