Iterative Refinement: Polishing Your Prompts
You will be able to diagnose exactly why an answer missed, give one precise correction at a time while protecting what already works, and capture a polished prompt as a reusable template instead of restarting from scratch.
{"contentFormat":"slides.v1","completion":{"requireAllSlides":true,"requireQuiz":true},"slides":[{"kind":"title","eyebrow":"Module 2: Practical Prompt Engineering","title":"Iterative Refinement: Polishing Your Prompts","body":"Professional prompt engineers rarely get perfect results on the first try. Instead, they treat prompting as a short conversation of adjustments. This lesson teaches you how to diagnose why an answer missed the mark, give targeted feedback, and turn a rough prompt into a polished, reusable one.","outcomes":["Explain why the first answer is rarely the final answer","Diagnose common issues in AI outputs (format, tone, length, etc.)","Apply iterative refinement in a conversation or by editing the prompt","Capture winning prompts as reusable templates"],"narration":"Welcome to Iterative Refinement. In this lesson, you'll learn that the first answer from an AI is just a starting point. By making small, precise adjustments, you can steer the model to exactly what you need. Let's get started."},{"kind":"content","heading":"What Is Iterative Refinement?","body":"Iterative refinement means improving your prompt or the model's output step by step, using each response to inform the next request. Rather than rewriting from scratch when an answer disappoints, you make a precise adjustment and try again.\n\nAnalogy: Think of adjusting a shower's temperature. You don't demand 'perfect' in one move — you nudge the tap, feel the result, and nudge again. A few small corrections get you exactly where you want. Prompting works the same way.\n\n### Two ways to iterate\n\n- Refine the prompt. Change the instruction itself — add missing context, tighten the format, adjust the tone — and re-run.\n- Refine within the conversation. Keep the current answer and give follow-up feedback: 'Good, but make paragraph two shorter and add a local example.' Because the model remembers earlier turns in the same chat, it builds on what's already there.\n\nUse in-conversation feedback for quick tweaks; update the underlying prompt when you want a clean, reusable version.","callout":{"variant":"note","title":"In-Conversation Refinement Example","text":"User: 'Write a short product description for a solar charger.'\nAI: [draft]\nUser: 'Good, but make it under 50 words and mention it works in cloudy weather.'\nAI: [revised draft]\nThis is a follow-up turn that refines the output without restarting."},"narration":"Iterative refinement is about making small, targeted changes rather than starting over. You can refine the prompt itself or give follow-up feedback within the same conversation. The model remembers context, so you can build on previous outputs."},{"kind":"content","heading":"Diagnose Before You Fix","body":"The key skill is naming exactly what's wrong. Ask yourself:\n\n- Wrong format? → Specify structure: 'Return a 3-column table.'\n- Wrong tone or level? → Name the audience: 'Rewrite for a beginner; friendlier tone.'\n- Wrong length? → Give a number: 'Cut to under 120 words.'\n- Missing information? → Add the context you left out.\n- Too generic? → Ask for specifics: 'Include a concrete example and real figures in euros.'\n- Went off-track? → Restate the core goal and the single most important constraint.\n\nVague feedback ('make it better') produces vague changes. Precise feedback produces precise changes.","table":{"headers":["Issue","Diagnosis Question","Example Fix"],"rows":[["Wrong format","Is the structure correct?","Return a 3-column table with headers: Product, Price, Rating."],["Wrong tone","Does the tone match the audience?","Rewrite for a beginner; use a friendly, encouraging tone."],["Wrong length","Is it too long or too short?","Cut to under 120 words."],["Missing info","Did I include all necessary context?","Add that the target market is small businesses in Nigeria."],["Too generic","Does it lack specifics?","Include a concrete example with real figures in euros."],["Off-track","Did it address the core goal?","Restate: the goal is to persuade investors, so focus on ROI."]]},"narration":"Before you fix, diagnose. Identify the exact issue: format, tone, length, missing context, genericness, or off-track. Precise feedback leads to precise improvements."},{"kind":"content","heading":"Give Feedback Like a Coach","body":"When refining, use these techniques:\n\n- Keep what works. 'The structure is perfect — keep it. Only change the opening.'\n- Change one thing at a time. If you alter five things at once and it improves, you won't know which change mattered.\n- Show, don't just tell. If wording is off, paste a sentence rewritten your way as a mini-example (a one-shot signal) so the model matches your voice.\n- Ask the model to help diagnose. 'Before rewriting, list 3 reasons this draft might not suit a non-technical reader.' Then have it apply its own fixes.\n\n### Capture the Winning Version\n\nOnce you've refined a prompt into something that reliably works, save it as a template with bracketed placeholders:\n\n
\nRole: You are a [profession] writing for [audience].\nTask: [what you want].\nFormat: [structure and length].\nConstraints: [tone, language level, must-keep facts].\n\nWrite a LinkedIn post about learning AI skills.\n\nGood start. Now: cut it to under 100 words, use a warmer first-person tone, and open with a question. Keep the hashtags.\n[brackets] and save it. You've just built a reusable asset.","callout":{"variant":"exercise","title":"Your Turn","text":"Try the exercise now with a chatbot. After step 5, reflect: how many iterations did it take? What was the most effective correction?"},"narration":"Now it's your turn. Follow the five steps: start rough, diagnose, give one precise correction, iterate once more, and capture the template. This hands-on practice will solidify the skill."},{"kind":"quiz","heading":"Check Your Understanding","questions":[{"question":"What is the main benefit of iterative refinement over starting from scratch?","options":["It guarantees a perfect answer on the first try.","It saves time by building on what already works.","It requires no feedback from the user.","It only works with paid AI models."],"questionId":"cmrf73k4j000opd278sgte0tb"},{"question":"Which of the following is an example of precise feedback?","options":["\"Make it better.\"","\"Cut to under 100 words and use a friendly tone.\"","\"This is not good.\"","\"Rewrite the whole thing.\""],"questionId":"cmrf73k4j000ppd27by2ehae7"},{"question":"Why should you change only one thing at a time when refining?","options":["It's faster to change many things at once.","You can identify which change caused the improvement.","The model can only handle one change per turn.","It prevents the model from making errors."],"questionId":"cmrf73k4j000qpd27vow7jrbn"}],"narration":"Let's test your understanding with a quick quiz. Choose the best answer for each question.","quizId":"qz_cmk7lg7lg0009g4p8ytavgtcz"},{"kind":"summary","heading":"Key Takeaways","takeaways":["Iterative refinement means improving prompts and outputs step by step — the first answer is a draft, not a verdict.","You can refine the prompt itself or give follow-up feedback within the same conversation.","Diagnose precisely — format, tone, length, missing context, or too generic — before you correct.","Change one thing at a time and explicitly protect what works to avoid regressions.","Save your polished prompts as reusable templates with bracketed placeholders."],"narration":"To summarize: iterative refinement is a powerful habit. Diagnose precisely, change one thing at a time, and save your winning prompts as templates. This approach will save you time and produce consistently better results."}]}