Home / Techniques

Prompting Techniques

Great prompting isn't magic—it's a set of learnable techniques. Master these and you'll get dramatically better results from any AI model.

1. Chain-of-Thought Prompting

What it is: Asking the AI to show its work and reason step-by-step instead of jumping to conclusions.

Why it works: AI models (like humans) make fewer reasoning errors when they break problems into steps. Explicit reasoning also lets you spot where things go wrong.

Before

What should I do about my startup's declining user engagement?

Result: Generic advice like "improve your product" and "talk to users."

After

My startup's user engagement has declined 30% over the past 3 months.

Walk me through diagnosing this step-by-step:
1. What data should I look at first to understand the scope?
2. What are the most likely root causes, and how would I test each?
3. For each likely cause, what would be the first validation step?
4. What metrics would tell me if my fix is working?

Think through each step before giving recommendations.

Result: Structured diagnostic process with specific, actionable steps.

When to use

Pro tip: Add "Let's think through this step by step" or "Show your reasoning" to almost any analytical prompt for better results.

2. Few-Shot Prompting

What it is: Giving the AI examples of what you want before asking for new output.

Why it works: AI learns patterns from examples better than from descriptions. Show, don't just tell.

Before

Classify these customer support tickets by urgency.

Result: Inconsistent classification based on the AI's assumptions about urgency.

After

Classify each ticket as HIGH, MEDIUM, or LOW urgency.

Examples:
Ticket: "The entire site is down and we're losing sales"
Urgency: HIGH

Ticket: "I'd like to request a feature for next quarter"
Urgency: LOW

Ticket: "My report exports are timing out"
Urgency: MEDIUM

Ticket: "The login button doesn't work on mobile"
Urgency: HIGH

Now classify these:
[TICKETS TO CLASSIFY]

Result: Consistent classification matching your criteria.

When to use

Pro tip: Include 3-5 diverse examples covering edge cases. More examples generally help, but quality matters more than quantity.

3. Role Assignment

What it is: Asking the AI to adopt a specific persona, expertise, or perspective.

Why it works: Role-playing activates relevant knowledge patterns in the model. A "marketing expert" responds differently than a "software engineer."

Before

How should I price my new SaaS product?

Result: Generic pricing advice.

After

You are a pricing strategist with 20 years of experience in B2B SaaS.

I've built a project management tool for remote engineering teams. Key details:
- Current competitors charge $10-25/user/month
- We have unique features around async collaboration
- Target customers are 50-200 person startups
- We're pre-revenue and need to balance growth with sustainability

How should I approach pricing? Consider:
- Pricing model (per seat, usage-based, tiered?)
- Price anchoring strategies
- When to grandfather early users
- How to test pricing before committing

Result: Expert-level strategic advice tailored to the situation.

When to use

Pro tip: Be specific. "Experienced divorce attorney" gets better results than "lawyer." Include relevant background in the role description.

4. Output Structuring

What it is: Explicitly defining the format, structure, and organization you want.

Why it works: AI defaults to paragraphs. If you want something else, you have to ask. Clear structure makes output more useful and easier to act on.

Before

Analyze our competitors.

Result: Unstructured paragraphs mixing everything together.

After

Analyze our top 3 competitors: [Competitor A], [Competitor B], [Competitor C].

For each competitor, provide:

OVERVIEW
- Founded: [year]
- Funding/Size: [info]
- Target market: [description]

STRENGTHS (2-3 specific, evidence-based)
WEAKNESSES (2-3 specific, evidence-based)
POSITIONING (how they describe themselves)
PRICING (what they charge, pricing model)

STRATEGIC THREAT LEVEL
- High/Medium/Low with justification

OPPORTUNITIES FOR US
- Gaps we could exploit
- Differentiation opportunities

Format as markdown. Be specific—avoid generic statements like "they have a strong team."

Result: Consistently structured analysis you can scan, compare, and act on.

When to use

Pro tip: Use specific markers (ALL CAPS, markdown headers, bullet points) to enforce structure. The AI will follow your formatting cues.

5. Iterative Refinement

What it is: Treating prompting as a conversation, not a single shot. Build on previous outputs.

Why it works: Complex tasks can't be fully specified upfront. Iteration lets you course-correct and add nuance.

Before

[One massive prompt trying to specify everything]

Result: Overwhelming, often misses the mark on key aspects.

After

Round 1: "Give me 10 blog post ideas about remote work."

Round 2: "I like ideas 3 and 7. Develop full outlines for both with H2s and key points."

Round 3: "Let's go with idea 3. Write the introduction and first section."

Round 4: "This is too formal. Make it more conversational and add a personal anecdote hook."

Round 5: "Better. Now add data points to support the claims in paragraph 3."

Result: Polished output shaped through feedback, not guessed upfront.

When to use

Pro tip: Save successful iteration patterns. They become reusable workflows.

6. Context Priming

What it is: Loading relevant context before making your request.

Why it works: AI has no memory of your situation. The more context you provide, the more tailored and useful the response.

Before

Should I pivot my startup?

Result: Generic advice that ignores your specific situation.

After

Context about my situation:
- Product: B2B SaaS for restaurant inventory management
- Traction: 12 customers, $2K MRR, launched 8 months ago
- Burn rate: $15K/month, 6 months runway remaining
- The problem: Customer acquisition is slow (3-month sales cycle), churn is high (15% monthly)
- The team: 2 founders (technical), no sales expertise
- Market feedback: Users love the product but say it's hard to justify cost for small restaurants

Based on this context, should we:
A) Double down on current market with better sales
B) Pivot to serve larger restaurant chains
C) Pivot to adjacent market (retail inventory)
D) Something else

What additional information would help you give better advice?

Result: Situation-specific analysis recognizing constraints and tradeoffs.

When to use

Pro tip: Structure context clearly (bullet points, sections) so the AI can parse it. Include constraints, goals, relevant history, and what you've already tried.

7. Constraint-Based Prompting

What it is: Explicitly stating what the output should NOT include or what limits to respect.

Why it works: AI defaults to comprehensive responses. Constraints force focus and prevent unwanted elements.

Before

Summarize this article.

Result: Summary that's too long, includes unnecessary details, misses your use case.

After

Summarize this article in exactly 3 sentences.

Constraints:
- Maximum 25 words per sentence
- Focus only on findings relevant to healthcare applications
- Do not mention the authors or their affiliations
- Use simple language (8th-grade reading level)
- No jargon without explanation

Article:
[PASTE ARTICLE]

Result: Tightly focused summary matching your exact specifications.

When to use

Pro tip: Phrase constraints as "Do not..." or "Avoid..." Negative instructions often work better than trying to specify everything you do want.

Putting It All Together

These techniques compound. A great prompt often uses several:

You are an experienced UX researcher (Role Assignment).

I'm showing you 3 examples of user feedback and how I categorized them (Few-Shot).

Now categorize these 10 new pieces of feedback using the same approach (Few-Shot continued).

For each, provide:
- Category
- Sentiment
- Priority
- Suggested action

Think through ambiguous cases step by step (Chain-of-Thought).

Do not include the original feedback text in your output—just the analysis (Constraint).

Quick Reference

Technique Use When Key Phrase
Chain-of-Thought Complex reasoning "Think step by step"
Few-Shot Pattern matching "Here are examples..."
Role Assignment Need expertise "You are a [expert]..."
Output Structuring Need organized output "Format as..."
Iterative Refinement Complex tasks Multi-turn conversation
Context Priming Situation matters "Context: [details]"
Constraint-Based Need limits "Do not..." "Maximum..."

Ready to apply these? Browse the Prompt Library for ready-to-use examples.