Skip to content

Basics

What Is a Prompt? Definition, Structure and Examples

A prompt is the instruction you give an AI. The clearer it is, the more useful the result. This article explains what makes a good prompt, which techniques help and which mistakes to avoid.

Lurus Editorial Team Published Reviewed 4 min read
Painted scene: an open notebook with sketches by a window overlooking a lake.

Key takeaways

  • A prompt is any input you use to give an AI a task.
  • Good prompts state the role, task, context, format and constraints.
  • Examples in the prompt (few-shot) and intermediate steps measurably improve results.
  • Confidential and personal data only belong in approved tools.
Contents
  1. Prompt: the definition
  2. The five building blocks of a good prompt
  3. Example: from a weak to a strong prompt
  4. Techniques for better results
  5. Common mistakes
  6. Sharing prompts across the team

Prompt: the definition

A prompt is the input you use to give an AI model a task – a question, an instruction or a text to be processed. The model then generates an answer.

Prompts can be short (“Summarize this text in three sentences”) or detailed, with background, examples and format requirements. Language models know nothing about your situation, your company or your customers unless you tell them. That is why the prompt largely determines the quality of the answer.

The five building blocks of a good prompt

A proven pattern consists of five building blocks. Not every prompt needs all of them, but the more important the result, the more they pay off.

  1. Role: From which perspective should the AI answer? “You are an experienced sales assistant in mechanical engineering.”
  2. Task: What exactly should it produce? “Write a reply to the following customer inquiry.”
  3. Context: What information does the AI need? Product data, audience, history, tone.
  4. Format: What should the result look like? “No more than 150 words, three paragraphs, formal tone.”
  5. Constraints: What should the AI avoid? “Do not mention prices or delivery dates.”

Example: from a weak to a strong prompt

Weak:

Write an email to a customer about the delay.

The AI knows neither the customer nor the product, the reason or the tone. The result will be generic and need heavy editing.

Strong:

You are a customer service representative at a mid-sized furniture manufacturer. Write an email to a business customer whose order of 40 office chairs is delayed by two weeks because a supplier delivered parts late. Tone: friendly, factual, solution-oriented. Offer to bring forward a partial delivery of 20 chairs. No more than 120 words. Do not offer discounts.

This prompt usually produces a draft that only needs to be checked and adjusted.

Techniques for better results

Provide examples (few-shot)

Show the AI one or more examples of the result you want. Research calls this few-shot prompting: large language models can solve new tasks from just a few examples in the prompt, without being retrained. This is especially useful for consistent formats such as product descriptions or ticket categories.

Ask for step-by-step reasoning

For complex tasks, it helps to explicitly ask the AI for intermediate steps: “Work step by step and justify each decision.” Studies on so-called chain-of-thought prompting show that intermediate steps improve the reasoning of large models on multi-step tasks. Many current models now do this on their own.

Follow up and refine

A prompt doesn’t have to be perfect the first time. Work in a dialogue: “Shorter”, “More formal”, “Add an example from logistics”. You can also ask the AI to ask you questions before it answers.

Ask for sources and uncertainty

Ask the AI to flag assumptions and say when information is missing. That doesn’t replace checking, but it makes weak points visible. Why this matters is explained in AI hallucinations.

Common mistakes

  • Too little context: The AI guesses what you mean – and often guesses wrong.
  • Several tasks in one sentence: Break complex assignments into separate steps.
  • No format: Without guidance, texts end up too long or poorly structured.
  • Blind trust: Even a perfect prompt doesn’t guarantee an error-free answer. You must check facts, figures and quotes.
  • Sensitive data in unvetted tools: What you enter leaves your company. Follow the rules of your AI policy.

Sharing prompts across the team

The biggest lever is often not the individual prompt but sharing: when a team collects proven prompts for recurring tasks, everyone benefits. Many tools also let you store instructions permanently as assistants. Ready-made templates for typical tasks are in our prompt templates, and more terms are in the AI glossary.

Frequently asked questions

Do I need to learn prompt engineering?

For everyday work, a few basic rules are enough: state the task clearly, provide context, define the format and check the result. Anyone who works with AI regularly benefits from a short training session and a shared collection of proven prompts.

How long should a prompt be?

As long as necessary and as short as possible. Simple questions need one sentence. For texts with fixed requirements, several paragraphs with context and examples make sense.

What is a system prompt?

A system prompt is a base instruction that applies before every conversation – such as tone, role or rules. Many tools let you store such instructions permanently for assistants.

May I put customer data in a prompt?

Only in a tool your company has approved for that purpose, with a data processing agreement and clear rules. Where possible, leave out names and identifiers or replace them with placeholders.

Sources

  1. Language Models are Few-Shot Learners (Brown et al., 2020) arXiv
  2. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (Wei et al., 2022) arXiv
  3. Generative AI Models – Opportunities and Risks for Industry and Authorities German Federal Office for Information Security (BSI)

Written by the Lurus Editorial Team according to our source standard. Editorial policy

Free initial consultation

Your AI start, planned together.

Show us your tasks. We’ll discuss data paths, rules and the right starting point for your team – no feature tour.

The initial consultation is free of charge. A phone number is not required.