Prompt Engineering Guide: Get Better Answers From AI in Any Situation

"Prompt engineering" sounds technical, but it's really just the art of writing clear, specific instructions for AI. The gap between mediocre and great AI output almost always comes down to how the question was asked. Better prompts mean better answers, less editing, and more time saved.

This guide walks through ten practical prompting techniques with plain examples, reusable templates, and common mistakes to avoid.

The Core Idea

AI responds to what you give it. Give it vague instructions and it will fill gaps with generic answers. Give it context, constraints, and examples, and it will match your intent much more closely.

• Vague: "Write me an email."
• Better: "Write a friendly follow-up email under 100 words to a client who missed our Monday meeting. Suggest 3 new times this week."

Technique 1: Give Context

Tell the AI who you are, who the output is for, and the situation.

Before: "Write a marketing post."

After: "I run a small bakery in Portland. Write a warm, playful Instagram caption announcing our fall menu to loyal local customers."

Technique 2: Specify the Format

Ask for exactly the format you want.

• Bulleted list
• Table with columns
• Word count or paragraph limit
• Numbered steps
• Email with subject line
• JSON, markdown, or plain text

Example: "Return as a 3-column table: Meal | Prep Time | Main Protein. Give 8 rows."

Technique 3: Set a Role

Give the AI a role to help it choose tone, depth, and jargon level.

• "You are a friendly customer service rep for a hardware store."
• "You are a senior editor at a business magazine."
• "You are a nutritionist explaining protein to a general audience."
• "You are an experienced technical writer producing SOPs."

Technique 4: Ask for Options, Then Choose

Instead of asking for one answer, ask for many.

• "Give me 10 subject lines for this email."
• "Suggest 5 different tones and pick one after seeing them."
• "Draft 3 versions: short, medium, and long."

Then pick, mix, or ask AI to combine the best parts.

Technique 5: Provide Examples (Few-Shot Prompting)

Show, don't just tell. Include 1–3 examples of the input and desired output.

Example:

Turn support tickets into 1-sentence internal summaries. Examples:

Input: "Customer says they can't log in after resetting password."

Output: "Login issue after password reset."

Input: "My subscription auto-renewed but I canceled last month."

Output: "Auto-renewal after cancellation."

Now do this one: "I got charged twice for the same order."

Technique 6: Iterate

The first answer is a draft. Ask follow-ups:

• "Make this shorter."
• "Now write it in a more formal tone."
• "Cut two paragraphs and tighten the intro."
• "Add three specific examples."
• "Rewrite as if you're speaking to a small business owner, not a corporate executive."

Technique 7: Ask AI to Critique Its Own Output

After AI gives you an answer, ask:

• "What's weak about this?"
• "What did you miss?"
• "What are 3 risks I'm not seeing?"
• "Where might this be inaccurate?"

Then have it improve based on its own feedback.

Technique 8: Constrain the Output

Give explicit boundaries.

• "Under 80 words."
• "No jargon."
• "Only use active voice."
• "Don't include statistics without a citation."
• "Include exactly 3 examples."

Technique 9: Chain Steps

Break big requests into smaller steps.

1. "Summarize this article in 5 bullets."
2. "From those bullets, draft an outline for a blog post."
3. "Now write the intro paragraph."
4. "Write section 2."
5. "Give me 5 title options."

Chained prompts produce far better results than one giant request.

Technique 10: Save and Reuse Your Best Prompts

• Keep a note file called "Prompt Library"
• Save prompts that worked really well
• Label them by workflow (marketing email, meeting summary, brainstorm)
• Share with your team or family
• Refine over time

Reusable Prompt Templates

The Explain-Like-I'm-New Template

"Explain [topic] to me like I'm [beginner / intermediate / expert]. Use analogies and short sentences. Then give me 3 follow-up questions I should ask to learn more."

The Draft-and-Critique Template

"You are a [role]. Draft [output]. Then critique your own draft — what's weak, what's missing, what could be tighter. Rewrite based on your critique."

The Multi-Option Template

"Give me 5 options for [goal]. For each: a 1-line summary, a strength, and a risk. Then recommend the best fit for [context]."

The Voice-Match Template

"Here is a sample of my writing: [paste]. Match this voice, rhythm, and tone in your response. Task: [describe]."

The Interview Template

"Interview me one question at a time to help me [goal — e.g., write a bio, plan my week, prepare for a talk]. Wait for each answer before asking the next question."

The Meeting Prep Template

"I have a meeting about [topic] with [person / role]. Draft: 3 goals, 5 questions to ask, 3 key talking points, and 2 possible objections. Context: [paste]."

What NOT to Do

• Ask vague questions ("Make it better")
• Send one long prompt with 10 unrelated tasks
• Assume the first answer is the best
• Trust facts without verifying
• Paste sensitive personal or business data unnecessarily
• Give up if the first prompt doesn't work — iterate
• Prompt in ALL CAPS or with aggressive language (rarely helps)

Choosing the Right Model

• Chatty, everyday tasks: Any modern LLM works
• Long reasoning, complex analysis: Use the strongest model available
• Coding: Coding-specialized models (Copilot, Cursor) or a strong LLM
• Speed and cost: Smaller/faster models for high-volume simple tasks
• Image, audio, video: Specialized generative tools

Testing Your Prompts

1. Try the prompt on a real task
2. Note what's good and what's off
3. Change one variable at a time
4. Save the winning version
5. Test on a second, similar task
6. Add it to your prompt library

Common Prompt Engineering Mistakes

• No role, no context, no format = vague answers
• Trying to do everything in one prompt
• Not iterating — first answer treated as final
• Assuming AI knows your business, brand, or audience
• Ignoring examples — few-shot prompting is high-leverage
• Not saving winning prompts
• Ignoring the model's limits (real-time info, exact math)

References

• National Institute of Standards and Technology (NIST) — AI Risk Management Framework.
• OpenAI — Prompt engineering guidance for developers and consumers.
• Anthropic — Claude documentation and prompt best practices.
• Stanford Human-Centered AI (HAI) — Prompt design research.
• Google DeepMind — Public papers on instruction tuning.

The Bottom Line

Better prompts equal better AI. Give context, specify format, provide examples, ask for options, iterate, and save your best prompts. Ten minutes learning these techniques will save you hours every week — for the rest of your working life. Start now, save what works, and share what you learn.