Writing Better Prompts
5 Tips for Better Results with ChatGPT, Copilot, and Claude
The Short Answer
Here are five tips for creating effective prompts: Structure the prompt around the role, context, objective, and format. Write in natural language, refine the result through dialogue, and critically evaluate each response. For complex tasks, advanced techniques such as the chain-of-thought method can be helpful. These prompting tips apply regardless of the tool—that is, they work equally well for ChatGPT, Microsoft Copilot, and Claude.
According to a Bitkom survey conducted in September 2026, 57 percent of companies in Germany use AI. At the same time, 66 percent rate their employees’ AI skills as low. In day-to-day work, this leads to vague prompts and mediocre results that require a lot of rework.
This article compiles five prompting tips that work with any AI tool. You’ll receive a guide with four building blocks, a before-and-after example, and an overview of advanced techniques.
5 Key Takeaways
- Good prompts deliver measurable results: In a Harvard study, consultants working with AI were 25 percent faster and delivered higher-quality work.
- Four building blocks form the foundation: role, context, objective, and format/scope turn a vague request into a clear assignment.
- Natural language works best: Write in complete sentences, just as you would when speaking to a new colleague.
- The first result is a draft: Refine it specifically through dialogue.
- Hallucinations are still possible: Good prompts reduce the risk, but they don’t replace your own review.
Why is good prompting so important?
12.2%
More tasks completed
out of 18 tasks from everyday consulting work
25.1%
Faster processing
compared to the group without AI
over40%
Higher quality
than the control group without AI
The quality of an AI response depends directly on the quality of the prompt. A study by Harvard and the Boston Consulting Group demonstrates just how significant this factor is: For 18 consulting tasks that GPT-4 handles well, consultants using AI completed 12.2 percent more tasks. They worked 25.1 percent faster and delivered over 40 percent higher quality than the control group without AI.
The basic rules of prompt engineering apply regardless of the tool. ChatGPT, Microsoft Copilot, and Claude are based on large language models and require the same input to deliver good results. That’s why those who are skilled at prompting can apply this ability to any new tool. Anthropic, the provider of Claude, compares the AI in its prompting recommendations to brilliant new employees who aren’t yet familiar with your workflows. The more precisely you explain what you want, the better the result will be.
Tip 1: Structure prompts according to the guide
A good prompt consists of four components. The ROOVER Prompt Engineering Guide helps ensure you don’t overlook any important details. Often, a single sentence is all it takes to quickly rephrase a prompt. The more important the result, the more worthwhile it is to follow the complete structure.
Click on a building block to see its effect and an example.
From what perspective should the AI respond?
Places the AI in the role of an expert. Its subject matter expertise, word choice, and tone are tailored accordingly.
“Act as an experienced HR manager at a medium-sized company.”
What does the AI need to know about the situation?
Provides background and context. The more the AI knows, the better the response will be tailored to you.
“We’re looking for a team leader for our customer service team of twelve employees in Cologne.”
What is the ultimate goal?
Tailor the response to the result you need. The AI knows what it’s working toward.
“Create a job posting that appeals to experienced executives who aren’t currently actively looking for a job.”
What should the result look like?
Specifies the length and structure. This ensures the result fits into your document without any additional work.
“A maximum of 250 words in the formal style, divided into tasks, profile, and offer.”
How does a good prompt differ from a bad one?
A good prompt provides the AI with all the information it would otherwise have to guess. An example from sales illustrates just how important the prompt is:
Weak prompt
“Write an email to our customers regarding the price increase.”
Strong Prompt
“1Role: Act as a customer service representative at a software provider for small businesses in the skilled trades. 2Context: As of January 1, 2027, our license prices will increase by 8 percent because we are expanding our phone support hours to 7 a.m. to 8 p.m. Many customers have been with us for years. 3Objective: Ensure that customers understand the reason and remain loyal to us. 4Format/Length: Write a maximum of 150 words using the formal “Sie” form, in a factual and respectful tone, with a subject line and contact person listed at the end.”
A weak prompt leaves it unclear who is writing, why prices are rising, and what tone customers expect. The AI fills in these gaps with generic phrases. A strong prompt produces an email that you just need to review and tweak.
How can your team learn to give better prompts?
Tip 2: Write in natural language
When writing with AI, write the way you would speak to a person. Imagine the AI as the most motivated intern you’ve ever had: it learns quickly and knows a lot, but it knows neither your company nor your expectations. You would speak to it in complete sentences. On the other hand, using only keywords—as you would in a Google search—leaves the AI with plenty of room to guess.
The reason lies in the training. Language models learn from large amounts of publicly available text, practicing predicting the next word in each case. Natural language therefore corresponds exactly to the material they were trained on. That’s why the AI also recognizes grammar, punctuation, and a polite tone. It takes these cues into account in its response. It ignores spelling errors, however, as long as the meaning of a word remains unchanged.
Also, describe the desired result and briefly explain your guidelines. Write “in continuous text with short paragraphs” instead of “no bullet points.” With an explanation, the AI will apply the rule to similar cases as well.
Tip 3: Improve Results Through Dialogue
The first result is rarely the final version. Treat it as a draft and refine it in the same chat. There, the AI is familiar with the conversation so far and can build on it. Here’s how to proceed:
- Provide specific feedback: Ask for specific changes, such as “shorter,” “more formal,” or “with two practical examples.”
- Provide examples: If the tone or structure isn’t quite right yet, attach a sample text. The AI will adopt its style directly.
- Work in stages: For extensive tasks, have the AI create an outline first. The individual sections are then developed step by step.
It’s only worth starting a new chat when the topic changes. Otherwise, you’ll lose the context you’ve already established.
Tip 4: Critically evaluate the results
AI models can also generate false information fluently and convincingly. In the case of a human, we would call this a lie; with AI, we refer to it as a hallucination. There is no intent behind it. The cause lies in how the technology works: A language model calculates the most likely continuation of a text. Since it has no consciousness, it cannot reliably assess how certain it is about an answer. Good prompts reduce the risk, but it cannot be completely eliminated.
Including the following information in the prompt reduces the risk:
- For each statement, provide a source or a quote from your documents.
- Limit the AI to the material you provide.
Still, verify numbers, names, and sources yourself before sharing any results. Depending on your company’s internal AI usage policy, you may be responsible for verifying AI-generated content. You’ll also be held accountable for any errors. Confidential data should only be entered into tools that your company has approved, such as ChatGPT Enterprise or Microsoft 365 Copilot.
Tip 5: Use advanced prompting techniques
For complex tasks, there are techniques that go beyond individual commands. These four are also useful for everyday office work:
Chain-of-Thought Prompting
The AI solves a task using transparent intermediate steps. Reasoning models already do this internally; in those cases, it’s often enough to simply ask them to think thoroughly.
For calculations and analyses
Tree-of-Thoughts Prompting
The AI develops multiple solution paths and evaluates them. It then pursues the best one. In tests, the solution rate for a math puzzle rose from 4 to 74 percent.
For strategic questions with options
Reflection Prompting
The AI checks its own answer for errors and then revises it. In tests, the accuracy rate for programming tasks rose from 80 to 91 percent.
For concepts and code
Interview Prompting
The AI will first ask you questions and will not provide the result until all the information has been provided. Example: “Ask me five questions one after another, and only then write the concept.”
For unclear requirements
You can find detailed examples of these techniques in our Prompting Guide for Businesses. We’ll show you how your team can apply them in their day-to-day work during the AI training session.
Frequently Asked Questions About Prompting Tips
FAQ on Prompting Tips
Long enough that the AI doesn’t have to guess. A single sentence is sufficient for a simple rephrasing. For texts, analyses, or concepts, five to ten sentences covering the purpose, context, objectives, and format are recommended. It’s best to attach long background texts as a separate document.
Yes, the basics apply to all major language models. The differences lie in the features: Microsoft 365 Copilot accesses emails, files, and appointments in your Microsoft environment. With ChatGPT and Claude, you usually upload the material yourself. You can read more about Copilot in the article on Copilot training in the workplace.
Not for most everyday work tasks. Current models understand German very well and respond in the language of the prompt. English can be helpful when it comes to technical terms from IT or research. Clear specifications regarding role, context, objectives, and format are more effective.
“Prompting” refers to the process of formulating instructions in everyday work. “Prompt engineering” goes a step further: Experts systematically test and refine prompts, for example, for chatbots or AI agents in automated workflows. For most employees, a solid understanding of prompting—as covered in this article—is sufficient.
Questions about prompting?
Oliver Breucker will answer your questions about prompting in your company.
Sources
- Bitkom: For the First Time, a Majority of Companies Are Using AI (September 14, 2026)
- Dell’Acqua et al.: Navigating the Jagged Technological Frontier, Harvard Business School (2023)
- OpenAI: How ChatGPT and Our Foundation Models Are Developed
- Anthropic: Best Practices for Prompting
- Yao et al.: Tree of Thoughts: Deliberate Problem Solving with Large Language Models (2023)
- Shinn et al.: Reflection: Language Agents with Verbal Reinforcement Learning (2023)
Researched with AI assistance, editorially reviewed by Oliver Breucker.
As of November 10, 2026.
About the Author
Oliver Breucker
Managing Director ROOVER
Oliver Breucker ist Experte für KI-Transformation bei Roover und ehemaliger Strategieberater von Roland Berger. Mit Expertise vom MIT unterstützt er Unternehmen bei KI-Strategien und digitalem Erfolg.
You might also be interested in
This week’s AI news (Week 46) begins with an unusual scene: Nvidia releases its quarterly earnings, presented by interim CEO Colette Kress instead of Jensen Huang. In addition, users can now pay AI agents in ChatGPT, and a Bitkom survey reveals the extent of concerns about uncontrollable AI. These news
AI governance in the company: The practical guide AI is already in use in many teams, but often without uniform rules. This is exactly where AI governance comes into play: who is allowed to do what, what data is permitted, how do we document decisions? Without this, it gets expensive: