The four parts of a clear request

Most disappointing answers are not a model problem. They are a briefing problem. Hand a competent freelancer the sentence “write something about our new pricing” and you’d get something unusable too — not because they’re bad at their job, but because you didn’t brief them.

Here is the briefing checklist. Four parts, fifteen seconds.

1. Task — what do you actually want done?

Use a verb. Summarise. Compare. Draft. Critique. Rewrite. List. “Something about our pricing” has no verb, so the model picks one for you, and it usually picks “explain” — which is rarely what anyone wanted.

Be wary of stacking five verbs into one request. “Summarise this, then critique it, then rewrite it, then make it shorter” gets you four things done at medium quality. Ask for one thing well, then continue.

2. Context — what should it work from?

The single highest-value thing you can add, and the subject of the next lesson. Paste the actual email. Attach the actual policy. Name the real constraint: the budget, the deadline, the word count, the thing you’ve already tried and rejected.

Without this, the model has only its general knowledge — which is exactly how you get advice that is true in general and useless to you in particular.

3. Format — what shape should the answer be?

Bullets or prose. Two hundred words or two thousand. A table with these three columns. An email ready to send. A list of questions, not answers.

Unstated format is the most common cause of “it gave me an essay when I wanted a list”. The model isn’t guessing badly; you left the field blank and it filled it with the most common shape.

4. Audience — who is going to read this?

The quietly powerful one. “Explain our refund policy” and “explain our refund policy to a frustrated customer who has already waited three weeks” are different jobs, and only one of them produces something you can send.

Audience sets vocabulary, assumed knowledge, tone and length all at once. It’s one clause and it does four jobs.

Watch it work

Before: “Help me with my presentation.”

After: “Turn these five bullet points into speaker notes for a ten-minute talk (task + format). The audience is our regional sales team, who know the product but not the new pricing model (audience). Keep it under 900 words, conversational, and flag anywhere I’m making a claim I should have data for (context + constraint).”

Same person, same tool, same five bullets. The second one comes back usable.

Try it now (5 minutes)

Take the weakest prompt you wrote in L1. Run the checklist over it and mark which of the four parts is missing — usually it’s two of them. Add only the missing parts, send it again, and notice that you didn’t need any clever phrasing at all.

Check your understanding

1. The four parts of a clear request are: Task, context, format, audience Role, tone, length, deadline Question, answer, check, repeat

2. You keep getting long essays when you wanted a short list. The missing part is most likely: Task Format Audience

3. Why is naming the audience unusually efficient? It makes the model try harder One clause sets vocabulary, assumed knowledge, tone and length at once It unlocks a specialised model

Recap

Task, context, format, audience — a fifteen-second checklist that fixes most weak prompts before sending. Audience does the most work per word. Next: why context is the part worth the most, and how to supply it well.