Tool · runs in your browser

The prompt workshop: build it, don't copy it

Good instructions consist of six building blocks. Choose what you want to do. Your selection becomes the instruction, block by block, with the reasoning for each. If you already have a prompt, check it here: which blocks are missing, and is there data in it that does not belong there? Everything in your browser, nothing leaves it.
What do you want to do?
Format
Length
Tone

For advanced users

Your instruction

Copy it, replace the placeholders in [BRACKETS], try it out

  1. 1 · Role and context

    You are helping me as an experienced, calm customer service agent of an Austrian business.

    Why: the AI knows nothing about you. Stating role and context up front saves you corrections afterwards.

  2. 2 · Task

    Draft a reply to the customer email below. The customer is annoyed; we want to solve the matter without making any legal commitment.

    Why: one task per instruction – several steps of the same task are fine, two different goals are not.

  3. 3 · Material

    Here is the material, clearly separated: """ [PASTE THE CUSTOMER EMAIL HERE] """

    It usually works just as well without real names: replace personal and contract details with placeholders like “customer A” – it almost never changes the result. The data traffic light shows what may go in.

    Why: separate the material cleanly – the AI has to tell instruction and text apart.

  4. 4 · Format and length

    Answer in prose. As a guide: no more than 120 words.

    Why: if you do not set the format, you get the AI's favourite format – not your own.

  5. 5 · Tone

    Tone: matter-of-fact and brief.

    Why: tone can be set like a typeface. But only if you ask for it.

  6. 6 · Quality rules

    Invent nothing. If you are missing information, ask me first instead of guessing.

    Why: the most important block of all. It turns confident guessing into honest questions far more often.

Remember: you are not talking to a person, you are configuring a machine. Precision is not rudeness.

Your input stays in your browser; this tool transmits nothing and stores nothing. The link to your selection contains only numbers, no text.

Already have a prompt? Check it.

Paste an instruction you use today. The check shows which of the six building blocks are in it and which are missing. It also flags personal or contract data in the text that should be replaced by placeholders before it goes into an AI tool. One click does either.

No text yet. Paste an instruction.

    The check works with simple patterns and only in your browser: your text is neither transmitted nor stored. It detects email addresses, phone numbers, IBANs, dates, salutations with names and street addresses, but not every piece of personal data. What may go into an AI tool is your decision. The data traffic light shows what may go in.

    What is a prompt?

    A prompt is the work instruction you give an AI. The more precisely role, task, material, format and limits are described, the more usable the answer. Not because the AI “understands better”, but because less has to be guessed.

    Why are copy-and-paste prompt templates not enough?

    Templates never fit your case exactly. Once you have understood the six building blocks, you build every instruction yourself. That is what this workshop is for. The templates above are illustrations, not a list to work through.

    What is the most common mistake?

    Missing quality rules. Without “Invent nothing, ask instead”, the AI delivers a confident answer even where its knowledge runs out. That single line turns invented details into honest questions far more often. This is why block 6 cannot be switched off in this workshop.

    What does the prompt check do with an existing prompt?

    It reads the text with simple patterns and reports which of the six building blocks it finds. Whatever is missing it appends as placeholders on request, so you fill in instead of starting over. A data radar also checks for email addresses, phone numbers, IBANs, dates, salutations with names and street addresses and replaces them on click with placeholders such as [NAME] or [IBAN]. Both happen in your browser; the text never leaves it.

    Frequently asked questions about the prompt workshop

    Does this work with every AI tool?
    Yes. The six building blocks are independent of the provider. The building blocks describe what an instruction has to contain, not which product reads it. Differences between tools concern length and formatting rather than structure.
    Is my input stored?
    No. This tool transmits and stores nothing. No. The instruction is assembled in your browser from text blocks delivered with the page; the prompt check reads your text there as well. There is no server request, and this tool stores nothing in the browser. The link to your selection contains only the numbers of the chosen template and options, no text. The website stores only your theme choice if you switch it. Opening the page itself produces server log data, as on any other website.
    May I put customer data into the placeholders?
    For this tool yes – for the AI tool afterwards it depends. You then paste the finished instruction into an AI tool, and there the kind of data decides what is permissible. The data traffic light shows how each combination of data and tool is to be assessed.
    How long may an instruction be?
    As long as necessary – but every line must do a job. Long instructions rarely fail because of their length, but because of contradictions: two jobs, two tone settings, two formats. If you keep the blocks apart, you can shorten or sharpen each one on its own.
    What does the data radar detect, and what not?
    Typical patterns such as email, phone, IBAN, date, salutation with name and street addresses. Names without salutation, company names or diagnoses it does not detect. The radar works with patterns, not understanding. It finds what has a recognisable shape. It misses what is personal only in context: a first name in a sentence, the company name of a sole trader, an illness, a dispute. Treat the result as a reminder, not as clearance. Which type of data may go into which AI tool is shown by the data traffic light; responsibility for what you paste stays with you.

    You have the building blocks. Now for your own case?

    In the use-case workshop your team works out the applications where the effort pays off. The first conversation before that is free.

    Arrange a call