8 Steps Prompt Framework

Guide: how do you write a good prompt?

A good prompt is the key to using AI effectively. Countless theories and formulas for prompts exist. But it all starts with thinking logically.

If you ask a language model simple questions, you do not need a complicated prompt. But if you want to get everything out of prompting, these frameworks are truly valuable!

That is why we use the 8 step Prompt framework (ROC-B-CPF-E): 8 steps that provide clarity, structure and usable output. It is derived from the shorter 6 step prompt framework. Below is an explanation of our own extended framework: the 8 step prompt framework!

1. Role (Persona)

What does this do? You decide which role the AI answers from.

Why does it matter? A policy advisor, lawyer or communications specialist each looks at the same topic differently. The chosen role determines the tone and angle.

Examples:

• "You are an experienced policy advisor who can translate complex regulations into clear and practical insights."
• "You are a professional editor who makes complicated information understandable for a broad audience."

Tip: Choose the role that fits your group or end goal (strategy, communication, analysis, education, etc.).

2. Objective

What does this do? You briefly describe what the AI should deliver.

Why does it matter? A concrete goal prevents the AI from drifting or answering too broadly.

Examples:

• "The goal is to create a clear overview of relevant schemes or insights based on the supplied documents."
• "The goal is to write a draft document suitable for publication or internal decision-making."

3. Context

What does this do? You explain the background and the audience.

Why does it matter? Context helps AI select relevant details and use the right tone.

Examples:

• "The organization wants to inform its audience about relevant developments within this topic."
• "The audience consists of professionals and decision-makers. Use an accessible, businesslike tone."

4. Sources

What does this do? You specify which documents or data must be used.

Why does it matter? AI works better with explicit input — it makes the answer factual and verifiable.

Examples:

• "Use only information from the supplied documents."
• "Base your analysis and summary only on these sources and make assumptions explicit."

5. Constraints

What does this do? You set limits on tone, length, language and structure.

Why does it matter? Without constraints you often get unclear or overly long texts.

Examples:

• "Write a maximum of 200 words for the summary."
• "Use a neutral, professional writing style."
• "Create an overview table with fixed columns such as: topic | audience | goal | conditions | relevant details."

6. Process

What does this do? You describe the logical sequence of steps the AI must follow.

Why does it matter? A step-by-step plan prevents AI from writing randomly or skipping things.

Example:

1. Analyze the supplied documents.
2. Identify the most important relevant parts.
3. Create an overview with key information.
4. Condense this into a short summary of no more than 200 words.
5. Expand this further into several thematic paragraphs.

7. Format

What does this do? You decide what the output should look like.

Why does it matter? A fixed format ensures consistency between teams and makes reuse easier.

"Deliver the document in the following structure:"

• Summary (max. 200 words)
• Overview table
• Section 1 – Background and context
• Section 2 – Analysis or explanation
• Section 3 – Relevance for the audience
• Source attribution + transparency about assumptions

"Use clear headings (H1, H2, H3) and bullet points where possible."

8. Evaluation

What does this do? You ask AI to check the quality of the output and suggest improvements.

Why does it matter? This way AI can correct itself — which increases reliability.

Examples:

• "Check whether all supplied sources have been used."
• "Ask yourself: 'Is this output relevant and understandable for the audience?'"
• "End with the question: 'Would you like me to rewrite this for a different audience or use case?'"

Good luck writing your prompts!