Prompt engineering shows up everywhere. TÜV NORD even trains people to become a “Prompt Engineer (TÜV)” (in German), and dozens of frameworks and guides circulate online. For questions about your project plan, you need far less. Here you learn what a prompt is, why context matters more than any formula, how the RACE memory aid works, and how to ask the AI to improve your prompt itself. On top, you get a cheat sheet with sentence templates for common questions.
The video for this article
The video (in German) is embedded in the German version of this article.
What is a prompt?
A prompt is the message you write to an AI assistant. Prompt engineering means building this message so that the answer fits your need. In the early years of large language models, a lot depended on the exact wording. Current models also understand loosely worded questions well. One principle has remained, though.
The principle: the AI knows only what you give it
The more relevant context, the more precise the answer. The sample project Produktlaunch Phoenix shows this well. This question is short and sounds reasonable:
Prompt
How is the project going?Through the MCP server, Claude knows the whole plan and answers with a broad status overview. What Claude does not know: what you need the answer for and for whom. The MCP server supplies the plan; you supply the occasion.
Context can reach the AI in several ways: directly in the prompt, through uploaded documents, through saved instructions, or through an MCP server that reads data from another application. For questions about your plan, the MCP server covers most of it. Whatever the plan lacks, you add.
The memory aid: RACE
There are countless frameworks for structuring a prompt. One of the best known is RACE, named after Role, Action, Context, and Expectation:
| Element | The question behind it | Example |
|---|---|---|
| Role | In what capacity should the AI help? | You support me as the project manager of Produktlaunch Phoenix. |
| Action | What should it do? | Check the status of the marketing tasks. |
| Context | What is the occasion, what does it need to know? | Tomorrow I have an alignment meeting with the head of marketing, and the budget is tight. |
| Expectation | What should the result look like? | Three bullet points: what is on track, what is delayed, where I need a decision. |
Put together, this gives you the following prompt:
Prompt
You support me as the project manager of Produktlaunch Phoenix. Check the status of the marketing tasks. Tomorrow I have an alignment meeting with the head of marketing, and the budget is tight. Answer in three bullet points: what is on track, what is delayed, where I need a decision.The answer is shorter and fits the meeting you need it for. Claude read the plan in both cases; the difference comes from the details only you know.
If you want to remember only two elements, make them context and expectation. A role sometimes helps, but with current models it is rarely the deciding part.
Build your own prompt
Fill in the four RACE elements or insert a suggestion with one click. Your prompt takes shape below and is ready to copy.
Your prompt
Still empty.
Your prompt appears here as soon as you fill in an element.
If you want to remember only two elements: context and expectation. They change the answer the most.
Let the AI improve the prompt
You can hand part of the work to the AI. Write your question the way it occurs to you, and add one sentence:
Prompt
Before you answer, first improve this prompt and show it to me so I can review it.Claude then suggests a more detailed prompt. Read it, correct assumptions that are wrong, and add what is missing. Only then send it. This way the AI takes care of the structure, and you keep control of the content.
Cheat sheet: three sentence templates for common questions
To get started, three patterns help; you only insert your own terms:
Summarize: “Summarize [area]: [criterion 1] and [criterion 2].”
Example: “Summarize the marketing area: open tasks and owners.”
Compare: “Compare [A] with [B] and show the deviations.”
Example: “Compare the planned with the actual start date of all tasks in November.”
What if: “What happens to [target] if [variable] changes?”
Example: “What happens to the launch date if the beta release slips by one week?”
Add one sentence of context to each pattern, and the answers become more targeted.
Do you still need prompt engineering?
As a discipline of its own, less and less. Working with AI is becoming part of everyday media literacy, much like email or spreadsheets. What remains is the principle from this article: give the AI the context it cannot find on its own.
Continue with the prompt library
You find all prompts from this video series in our prompt library, which grows with every video. Five examples for everyday project work are in the article 5 questions for your project plan. You learn the most when you try it on your own project. A good starting point is the sample project Produktlaunch Phoenix. How to connect your AI assistant is shown in the article Which AI client is right for you?
If you have any questions about this blog article or would like to discuss it, we look forward to your contribution in our forum.
Frequently asked questions
What is a prompt?
A prompt is the message you write to an AI assistant. It can be a short question or a detailed instruction with context and the result you want.
Do I need a course in prompt engineering?
Not for questions about your project plan. Current models also understand loosely worded questions. More important than any formula is that you give the AI the context it does not know on its own.
What does RACE stand for?
RACE stands for Role, Action, Context, and Expectation. The memory aid helps you build a complete prompt.
Do I always have to use all four RACE elements?
No. If you want to remember only two, make them context and expectation. They change the answer the most. A role sometimes helps, but with current models it is rarely decisive.
Why does the AI answer “How is the project going?” in such general terms?
Through the MCP server, the AI knows your plan, but not your occasion. It does not know whether you need the answer for a meeting with management, the weekly report, or your own overview. So it answers broadly. One sentence of context makes the answer more targeted.
Can the AI improve my prompt itself?
Yes. Add a sentence to your question asking the AI to improve the prompt before answering and to show it to you. Review the suggestion, correct wrong assumptions, and add what is missing.
Do I have to write my prompts in English?
No. Current AI assistants understand prompts in many languages well and answer in the language you write in.
Do the prompts work with every AI assistant?
The prompts themselves do. For the AI to read your plan from Merlin Project, though, the assistant has to be connected to Merlin Project via the MCP server. The MCP documentation describes how.
