Talking to AI: prompting that works
Understand
"Prompting" just means how you ask. Because the model continues your text in the most likely way, the way you set up the request largely determines what you get back. This is the single highest-leverage skill in the whole course — small changes in the ask produce large changes in the output.
Four moves cover ~80% of the gain. You don't need clever tricks; you need to stop under-specifying.
- Give it a role and a goal. "You are a patient tax accountant explaining to a freelancer" beats no framing. The role pulls the answer toward the right region of everything it learned.
- Give it context. It can't see your situation. Who is this for, what do you already have, what's the constraint? "Rewrite this email" → "Rewrite this email to a client who missed a payment, friendly but firm, under 120 words."
- Say what 'good' looks like. Specify format, length, tone, audience. "A 5-bullet summary a non-technical manager could skim" is a target it can aim at. Vague in, vague out.
- Iterate — don't restart. Treat the first answer as a draft. "Good, but make it warmer and cut the jargon" is faster and better than rewriting the prompt from scratch. The conversation is the tool.
One more, free: show an example of what you want when you can. "Match the style of this sample: …" One example is worth a paragraph of description, because matching patterns is exactly what the model does best.
The meta-point: you are not 'searching' — you are briefing a collaborator. The quality of the brief sets the ceiling on the work.
See it
Weak prompt: "Give me marketing ideas." → a generic listicle that could be for anyone, useful to no one.
Strong prompt: "You're a growth marketer. I run a one-person ceramics studio selling £40 mugs, mostly to people aged 25–40 in the UK, no ad budget, ~800 Instagram followers. Give me 5 specific, low-cost experiments I could run this month. For each: the idea in one line, the first concrete step, and what result would tell me it's working."
Same model, same mechanism. The second prompt supplies role, context, and a definition of 'good' — so the likely continuation is now a tailored, actionable plan instead of filler. Nothing about the AI changed; the brief did.
Try it
Pick a real small task you have (an email, a plan, an explanation). First ask the AI for it in one lazy sentence. Then ask again using all four moves: role, context, what 'good' looks like, and one follow-up to refine. Compare the two outputs side by side.
The second result is visibly more useful and specific to you. You've directly felt that prompt quality — not the model — was the bottleneck, and you have a repeatable recipe (role · context · 'good' · iterate).
Tried it? Paste what you got and the tutor will tell you if it worked.
Key terms
- Prompt
- What you send the AI. Not a search query — think of it as briefing a collaborator. Its quality sets the ceiling on the output.
- Context
- The background you give the AI about your situation — who it's for, what you have, the constraints. It can't see any of this unless you say it.
- Role prompting
- Telling the AI who to be ('you are a patient tax accountant…'). It steers the answer toward the relevant region of what the model learned.
- Iteration
- Refining in the same conversation ('warmer, less jargon') instead of restarting. The conversation itself is the tool.
- Few-shot (example-giving)
- Showing the AI one or two examples of what you want. One example is worth a paragraph of description, because pattern-matching is what it does best.
All terms live in the keywords bank.
Explain it back
The real test: explain this in your own words, like you're teaching it. The tutor will tell you honestly if it's solid.
Check yourself
Your first answer from the AI is close but too formal. What's the most effective next step?
Why does adding a role and context ('You are X, helping Y, who needs Z') improve answers?