Homework — How to talk to AI well

Task 1 — The four-part prompt: a task that's real to you

Pick a task from your actual life or work — something you genuinely need done or have struggled to get good AI output on before. It could be drafting an email, planning an event, writing a bio, summarising a document, preparing for a conversation, or anything else.

Step A: Write a vague prompt first. Just a few words. Run it and note what you get.

Step B: Now write a four-part prompt for the same task. Use the recipe:

• Role: who the AI should act as.
• Context: the background it needs. What's the situation? Who's involved? What do you already know?
• Task: exactly what you want it to do. Be specific about scope and quantity.
• Format: how you want the answer. List? Table? Paragraph? Word count? Tone?

Run it. Compare the two responses.

Step C: Iterate at least twice in the same chat. Use one-line refinements ("shorter", "more formal", "add a section on X") to improve the output further.

Reflect and write down:

• What was the vague prompt? What did it produce?
• What was the four-part prompt? Paste it here.
• What changed? What was noticeably better in the second response?
• What two refinements did you make? Did the third or fourth version feel close to usable?

✓ Self-check: You should have two clearly different AI responses for the same underlying task, plus at least two refinement steps visible in your chat history.

Task 2 — Show an example and use negative constraints

This task focuses on Habit 1 (show an example) and Habit 3 (tell it what not to do) — two of the most powerful but least-used techniques.

The scenario: You need to write a short introduction for a speaker at an event (or adapt this to your own context: a product blurb, a job posting, a social-media bio — any short piece where style and tone matter).

Step A: Find or write one example of an introduction you like. It doesn't have to be long — two to four sentences is fine. This is your style reference.

Step B: Write a prompt that includes:

• The example you found (paste it in after "Write it in the style of this example:")
• Relevant context about the speaker (or product, role, etc.)
• At least two "don't" constraints, e.g. "Don't use jargon," "Don't start with 'I am pleased to introduce'," "Don't be longer than 80 words."

Run it and review the output.

Step C: Remove the example from your prompt and run it again with everything else the same. Compare.

Reflect and write down:

• Paste both prompts and both responses.
• Did the example meaningfully change the output's style? In what way?
• Did the "don't" constraints work? Were any ignored?
• What did you learn about how negative constraints interact with positive instructions?

✓ Self-check: You should have two responses — one with an example, one without — that differ noticeably in style or structure. You should also be able to point to at least one place where a "don't" constraint had a visible effect on the output.

Task 3 — The step-by-step technique on a tricky question

This task tests Habit 5: letting the AI think for hard problems by asking it to show its reasoning.

Pick one of these scenarios (or substitute your own similarly complex question):

• You're deciding between two job offers with different salaries, benefits, commutes, and growth potential. Ask the AI to help you think through the decision.
• You're planning a trip to a city you've never visited and want to optimise for 3 days, limited budget, and a specific interest (food, museums, street art — your choice).
• You're trying to understand a clause in a contract or rental agreement that confuses you. (Invent or paraphrase a clause if you don't have a real one.)

Step A: Write a prompt without "think step by step." Run it.

Step B: Add "Think step by step" (or "Show your reasoning") to the end of the same prompt. Run it.

Step C: Iterate once: ask a follow-up like "Now give me just the final recommendation in two sentences" to extract a clean conclusion from the reasoning.

Reflect and write down:

• What question did you choose, and why?
• Paste both versions of the prompt and their responses.
• Did the "step by step" version surface anything the first version missed? Did the structure of the reasoning feel more trustworthy?
• Was the final two-sentence summary useful? Did iterating in the same conversation feel natural?

✓ Self-check: You should have two responses to the same question — one with, one without step-by-step reasoning — plus a short follow-up refinement. The reasoning version should visibly break the problem into stages rather than jumping straight to a conclusion.

Task 4 (Bonus) — Build your personal prompt cheat sheet

After completing the three tasks above, you have enough experience to start building your own reference. This bonus task asks you to synthesise what you've learned into something you'll actually use.

Write a personal prompt cheat sheet. It should cover:

• The four-part recipe, with one worked example from your own life (not the recycling example from the lesson — use your own).
• Two or three "don't" constraints that are useful to you personally.
• One example you like that you could re-use as a style reference for future prompts.
• Any pattern you noticed in your homework that you want to remember.

Keep it short — half a page or a few bullet points. You can write it here, in a note-taking app, or on paper. The format doesn't matter. What matters is that it's yours.

✓ Self-check: Your cheat sheet should be something you'd actually open the next time you sit down to write a prompt. If it looks generic, make it more specific to how you actually work.