Prompt Engineering Basics

A prompt is any instruction, question, or context you give to an AI language model. The model processes your prompt by predicting the most likely continuation — the response — based on its training data and the specific words you provided.

This means two things matter above all else: specificity (what exactly you want) and context (what the model needs to know to deliver it). Vague prompts produce vague responses because the model fills ambiguity with the average of everything it’s seen.

The mental model: Think of the AI as an extremely capable contractor. The more clearly you specify the deliverable, the timeline, the constraints, and the quality bar — the better the result. “Build something” is a terrible brief. “Build a 3-bedroom house at 123 Main Street, 2,000 sq ft, under budget $X, completed by Y” is a good brief.

Every effective prompt contains some combination of these 5 elements:

1. Role: Tell the AI who it is. “You are an expert career coach with 15 years of experience in tech hiring.” Role-setting activates relevant knowledge and sets the tone.
2. Context: What does the AI need to know? Background information, the document being analyzed, your situation, constraints.
3. Task: What specifically do you want? Use action verbs: “Write,” “Summarize,” “Analyze,” “Compare,” “Extract.”
4. Format: How should the output be structured? “Return a numbered list,” “Write in bullet points,” “Format as a markdown table,” “Keep under 100 words.”
5. Examples: Show the AI what good looks like. Providing 2-3 examples of desired output dramatically improves consistency.

Example weak prompt: “Help me write a cover letter.”
Example strong prompt: “You are an expert career coach. Write a 250-word cover letter for [name] applying to [role] at [company]. Use a professional but conversational tone. Open with a specific connection to the company’s mission, highlight [top skill], and close with a clear call to action. Avoid phrases like ‘I am excited to apply.’”

For multi-step reasoning tasks, telling the AI to think through a problem step-by-step produces significantly better results than asking for a direct answer.

Add one of these to any complex prompt:

• “Think through this step by step before giving your final answer.”
• “First, analyze the problem. Then, consider three approaches. Finally, recommend the best one.”
• “Before answering, list your assumptions.”

Why it works: language models generate tokens sequentially — the “reasoning” they generate becomes additional context for subsequent tokens, improving logical coherence in the final output.

Use case example: Instead of “Should I accept this job offer?”, try: “I received a job offer. Here are the details: [details]. Analyze the pros and cons step by step, identify the top 3 risks, and give me a recommendation with reasoning.”

Few-shot prompting means including examples of the desired input-output pairs in your prompt. It’s the most reliable way to get consistent formatting, tone, and structure across multiple generations.

Template:

[Task description]

Example 1:
Input: [Example input]
Output: [Desired output]

Example 2:
Input: [Example input]
Output: [Desired output]

Now complete this:
Input: [Your actual input]
Output:

3-5 examples are typically optimal. Fewer leaves too much ambiguity; more increases prompt length with diminishing returns.

A system prompt is a persistent instruction set that shapes every response in a session or application. They’re the foundation of production AI tools — every AI assistant, chatbot, or automated workflow runs on a system prompt.

System prompt structure:

1. Identity: Who the AI is and what it’s optimized for
2. Capabilities: What it can and should do
3. Constraints: What it must not do
4. Format defaults: How it should structure responses by default
5. Tone and style: Voice, vocabulary, formality level

Example system prompt for a resume review assistant:

You are an expert ATS resume coach with deep knowledge of
Workday, Greenhouse, and Lever. When given a resume and job
description, you:
1. Score keyword alignment (0-100%)
2. Identify missing high-priority keywords
3. Suggest 3-5 specific rewrites for the most impactful bullets
4. Flag any ATS-incompatible formatting

Always be specific. Never give generic advice. Format your
response with clear sections and bullet points.

When a prompt doesn’t produce the desired output, diagnose before rewriting. Most failures fall into one of four categories:

Prompt versioning practice: Keep a running log of prompts that work well. The best prompt engineers treat prompts like code — versioned, documented, and shared with their team.