Before we dive into how to write effective prompts, it helps to pause and understand 'WHAT' you are working with and 'HOW' you are using the tool. In this part, we will: Look at a few simple truths about large language models (LLMs). Learn how to shift from getting mediocre results to good and great ones. Keep your "Prompt Away" task in mind as we go through this section.

You do not need to be a data scientist to use Copilot well. However, a few key ideas about LLMs will help you work with them more confidently and safely. LLMs cannot read your mind LLMs only work with what you tell them. If your request is vague, the output will be vague. LLMs will not proactively ask you questions You need to volunteer context, constraints, and missing details instead of waiting for the tool to ask. LLMs tend to be helpful and agree with you They may 'help' by building on a wrong premise. You can ask them to question your assumptions. LLMs can and will hallucinate Their answers come from patterns, not a source of truth. You are responsible for fact-checking. LLMs do not 'understand' like a person They predict likely words. You must apply your own judgment and adapt what you get. As you work with Copilot, keep these points in mind. They explain many of the frustrations people have with AI, and they also point to what you can do differently.

Now that you understand the foundation, let's put it into practice with a simple, repeatable framework. This 4-step process will help you move from a vague idea to a refined, useful output: Each step builds on the previous one. By the end, you will have a clear process you can reuse for any task. Step 1: Identify and Document Your Use Case Goal: Choose one real work task you want to improve with Copilot and document it clearly using a simple template. Input Your 'Prompt Away' task or real work example. Action Document your use case using a template. Output A clearly documented use case. Document Your Use Case Use Case Title Description How is this currently being done? How often and how long does this take now? What are the issues or challenges? How do you hope Copilot can help you? What does success look like? Any constraints or commentary? Don't have any idea? Use the "Idea Generator" Meta-Prompt You can use the following meta-prompt to give you some ideas on how it can help you. Note: To use this prompt well, you will need to set up your custom instructions in Copilot so it knows your role, goals, daily tasks, and challenges. Idea Generator Role: Act as an AI Productivity Consultant. Instructions: - Based on what you know about me, suggest the top 5 ways that you can help me in my context. - Keep all ideas within Copilot Chat's native capabilities (no external integrations). - For each use case, provide: Description, Expected Benefit, Ease of Implementation. Guidelines & Tips Be specific, not generic. "Write emails" is too broad. "Draft a weekly update email to my project sponsors based on my notes" is much more useful. Think about impact. Prefer tasks where saving time or improving quality will truly help you or your team. Choose something repeatable. A recurring task (weekly updates, regular reports, standard messages) gives you more return on the effort you invest today. Stay within policy. Avoid including highly sensitive or personal data in your examples unless it is clearly allowed by your organization's guidelines. Step 2: Craft Your Prompt Goal: Work with Copilot to turn your documented use case into a comprehensive prompt. Do not try to craft the prompt yourself. Let the LLM do the heavy lifting for you. Input Your documented use case from Step 1 Action Work with Copilot to create a comprehensive prompt Output A ready-to-use prompt Prompt Designer Act as an Expert Prompt Designer. Based on the use case details provided, draft a comprehensive and clear prompt I can use to ask for your help with this use case. The prompt should briefly restate the context, describe the output I want you to produce, include my key constraints, and end by flagging any assumptions and requesting any missing information. After you propose the prompt, suggest 2–3 optional tweaks I could make to adapt it for different situations. Here are the details: [PASTE YOUR DOCUMENTED USE CASE FROM STEP 1] Optionally run a quick “clarify and brainstorm” loop first if the use case is complex. Guidelines & Tips Let Copilot draft, you steer. Your job is to validate accuracy, tighten scope, and adjust tone. Avoid rewriting from scratch unless necessary. Sanity-check the working prompt before you run it. Confirm it clearly states: Goal: What you want to achieve Audience: Who it is for (and their expectations) Output: The format you want (email, bullets, table, checklist, etc.) Success: What “good” looks like (time saved, quality bar, completeness) Constraints: No-go areas, confidentiality limits, tools to use or avoid Make it easy to follow. Use short sentences, headings, and bullet instructions. Prefer explicit structure over long paragraphs. Be specific about inputs. If the output depends on your notes, data, or examples, say what you will provide and in what form (paste text, bullet notes, a draft, etc.). If something is unknown, tell Copilot what assumption to make. Optional: For more complex use cases A) Problem Clarification Problem Clarification Role: Act as a Problem-framing Partner. Task: Turn my documented use case into a clear problem statement and success criteria. Instructions: 1. Read my use case details below. Use only the information I provide. 2. If anything is missing or unclear, ask me up to 3 clarifying questions and wait for my answers. 3. Then respond with: * Concise Summary (≤ 50 words) * Problem Statement using this template: > "As a [role], I need help with [task / challenge] because [why it matters]. > Success means [outcome & any metrics]. > Key context: [contextual details, constraints, assumptions]." * What Success Looks Like: 3–5 clear, outcome-focused bullet points. Here are my use case details: [PASTE YOUR USE CASE FROM STEP 1] B) Brainstorming Brainstorming with Copilot Act as my AI Thinking Partner. Based on the use case details I share below, suggest several realistic ways Microsoft Copilot Chat could help me tackle this problem. Briefly explain the pros and cons of each option, recommend the 1–2 approaches that you think would work best in my context, and ask me any clarifying questions you need before confirming your recommendations. Here are the details: [Paste your problem statement, "what success looks like" bullets, and key constraints here] C) Then return to Prompt Designer After clarification/brainstorming, run the Prompt Designer prompt again, but paste your refine problem statement and success criteria. Step 3: Execute, Review and Refine Goal: Run your prompt with Copilot and improve the response through conversation until it is good enough to use in your real work. Input Ready-to-use prompt from Step 2 Action Execute the prompt and refine through conversation Output A refined, usable response for your real work Guidelines & Tips Use these mini-steps and questions to keep the conversation going: 1. Review Check relevance, completeness, and accuracy. Ask yourself: Does this answer my original question or solve the problem I described? Where are the gaps in this answer? What evidence or reasoning backs each key claim? 2. Clarify & expand Ask for deeper explanations, examples, and alternative views. Ask yourself: Which parts feel too high-level or generic to act on? Can I ask for specific examples, case studies, or analogies? How would an alternative view or scenario compare? 3. Feedback Tell Copilot what to keep, change, or remove. Ask yourself: Which parts hit the target, and which feel off-topic or weak? Is the level of detail right for my purpose and audience? Which points deserve more or less emphasis? Improve This Compared to my goals and success criteria above, here is what works well: [list] and here is what is missing, unclear, off-topic, or incorrect: [list] Please produce a revised version that keeps the strengths and fixes the issues. 4. Refine & iterate Decide whether to add context or adjust your prompt. Ask yourself: Is there important context I should add now? Should I narrow or widen the scope? Do I need one more cycle, or is this "good enough" for now? 5. Extend & reflect Ask Copilot to help with next steps. Ask yourself: What is the next step, and can Copilot help with that too? Can Copilot turn this into an action plan, checklist, or timeline? Extend Value Act as my AI Thinking Partner. Help me extend the value of this work by suggesting the most useful next steps (e.g. action plan, checklist, risk analysis) and how you can help me with these steps. Asking me clarification questions for additional context required. How Much of LLM's Potential Are You Using? Answering Machine Explainer & Tutor Creator & Transformer Brainstorm and Ideation Partner Planner & Strategist Devil's Advocate Role Play Partner Step 4: Learn and Grow (Use AI to Learn AI) Goal: Turn this conversation into learning so you can work even more effectively with Copilot next time. Input Your completed conversation and results Action Use Copilot to extract lessons and improve your prompting skills Output Learning insights to work more effectively next time Treat each interaction as a learning loop — not just about fixing the output but improving your prompting process. The goal is to get better at prompting by learning with AI. Use AI to Learn AI Act as my Prompting Coach. Looking at our full conversation above, review how I described my problem, how I provided context, and how I gave you feedback; tell me what I did well, suggest 3–5 specific improvements I can make next time, and summarize your advice as a short list of simple rules I can follow. Guidelines & Tips Treat this as part of the task, not extra work. A simple review can take just a few minutes and pays off across many future chats. Focus on what you can control. You cannot control how the model is built, but you can control: How clearly you define the problem How much useful context you provide How specifically you give feedback and corrections Check whether you played your role well. Ask yourself: Did I set a clear goal and success criteria? Did I provide enough context and constraints? Did I review and fact-check the output before using it? Look for patterns across different chats. Recurring habits (for example "I rarely mention my audience") are your best improvement opportunities. Keep your rules short and practical. Aim for 3–5 simple rules you can remember and actually use.