// knowledge · techniques

Prompt Engineering

How to write prompts that work — techniques, examples and templates from real-world practice

Basics

Why prompts matter

Prompt engineering is the art of phrasing a task so that AI understands exactly what you want — and delivers a result you can use straight away. It's not programming. It's communication.

What happens when you write a bad prompt: Bad prompt = bad output. AI doesn't guess your intent — it processes exactly what it gets. A vague brief returns a vague result. Missing context returns a generic answer. A missing output format returns chaos.

✗ Bad

Write me a strategy

✓ Good

You are an experienced strategy consultant. I work as a marketing director at a tech company. Propose 3 concrete steps to bring AI into a marketing team within 90 days. For each step: name, what exactly to do, expected outcome. Max 200 words.

Structure

What makes a good prompt

ROLE

Define who the AI should be. The more precise the description, the better the result.

"You are an experienced strategy consultant specialising in AI transformation at B2B companies."

Add relevant experience and context for the role — not just a title.

CONTEXT

What the AI needs to know about the situation, the company, the target audience.

"I work as a marketing director at a tech company, a team of 5, and we don't use any AI tools yet."

The more relevant context, the less generic the answer.

TASK

A precise description of what you want. One task = one prompt.

"Propose a 90-day plan for bringing AI into the marketing team."

Split complex tasks into smaller ones — one prompt, one output.

OUTPUT FORMAT

What the result should look like — structure, length, style.

"3 steps, each step: name + what to do + expected outcome. Max 200 words."

This is the most often forgotten part — and yet it's key.

VERIFICATION AND REASONING

Ask the AI to justify its conclusions or verify the data.

"For each step, state why you recommend it and what you base it on."

Prevents hallucinations and raises the quality of strategic outputs.

Techniques

Techniques that work

Technique 1

Zero-shot

You give the task without examples. Works for simple tasks where the intent is clear.

When to use: Translations, summaries, simple questions.

Summarise this article into 5 key points for a marketing manager: [ARTICLE]

Technique 2

Few-shot

You show the AI 2-3 examples of what the output should look like. Dramatically improves consistency.

When to use: When you want a specific format or tone that's hard to describe in words.

Write a LinkedIn post in this style: Example 1: [PASTE AN EXISTING POST] Example 2: [PASTE AN EXISTING POST] Topic of the new post: [TOPIC]

Technique 3

Chain-of-Thought

You ask the AI to think step by step. Ideal for analyses, brainstorming and writing briefs.

When to use: Strategy, competitor analysis, campaign briefs, complex decisions.

Analyse this situation step by step: 1. First summarise the key facts 2. Identify the main challenges 3. Propose 3 possible solutions with pros/cons 4. Recommend one solution and explain why Situation: [DESCRIPTION]

Technique 4

Role Prompting

You assign the AI a specific role. The more precise the role description, the better the result.

When to use: Always — role prompting should be part of every prompt.

You are an experienced marketing director who led marketing at a tech company for 10 years. You have experience with digital transformation and bringing AI into teams. You answer concisely and with a focus on action — no theory, just concrete steps.

Technique 5

System prompt

Persistent instructions that apply to the whole conversation. Set them once, they always apply.

When to use: When you use AI repeatedly for the same type of work.

You are an experienced strategy consultant specialising in marketing, innovation and AI transformation. In every answer: - Think in a structured way — clear steps, not general advice - If you're not sure about a fact, say so — don't make things up - Back every conclusion with reasoning or a source - Outputs are concise and actionable — no unnecessary waffle - For complex tasks, think step by step out loud Context: - The user works in marketing and innovation - Builds AI prototypes (vibe coding with Claude Code) - Is interested in GEO, agents and prompt engineering - Prefers outputs ready for direct use

Templates

Copy-paste templates

Content summary

Summarise the following [article/document/report] into a structured briefing for [TARGET AUDIENCE]. Format: - Key finding (1 sentence) - 5 main points (each max 2 sentences) - What it means for [AREA/TEAM] - Recommended next step Keep all important numbers and facts. Text: [PASTE CONTENT]

Step-by-step strategy

You are an experienced strategy consultant. Context: [DESCRIBE THE SITUATION AND GOAL] Propose a strategy step by step: 1. First analyse the situation — what we know, what we don't know 2. Identify 3 main challenges 3. Propose a concrete action plan — steps, responsibilities, timeline 4. For each step, list the risks and how to mitigate them 5. Define how we'll know it's working (KPIs) Be specific. Every step must be immediately usable.

Campaign brief

You are a senior brand strategist. Create a creative brief for [CAMPAIGN TYPE]. Fill in these sections: - Campaign goal (1 sentence) - Target audience (demographics + psychographics) - Key message (1 sentence people will remember) - Tone of voice - What the campaign must not say/do - Suggested formats - Measurable KPIs The campaign is for: [PRODUCT/SERVICE/COMPANY] Market context: [PASTE CONTEXT]

Vibe coding prompt

You are an expert in [TECHNOLOGY — Next.js/React/Python]. Build [WHAT YOU WANT TO BUILD]. Requirements: - [FEATURE 1] - [FEATURE 2] - [FEATURE 3] Design: [DESCRIBE THE STYLE or name an existing design system] Stack: [TECHNOLOGIES] Before writing code: 1. Summarise what you're going to build 2. Propose a file structure 3. Identify potential problems Then write complete, working code.

⚡

The biggest change in how I work with AI

It came when I understood the structure of a prompt and why it matters. The key moment was realising that I always have to describe the expected output — what it should look like and what it should contain. And that I have to ask the AI to verify the data and justify the result. That changed the quality of the outputs more than any other technique.

— Dave, daveai.cz

← Back to Knowledge