---
title: "The six-part prompt: role, context, task, constraints…"
description: "Prompts are briefs The biggest shift in prompting is to stop thinking of a prompt as a question and start thinking of it as a creative brief . If you…"
url: https://optimizeall.com/learn/prompt-engineering-for-content-and-sales/six-part-prompt-structure
updated: 2026-10-05
---

Prompt Engineering for Content & Sales · The anatomy of a great prompt · lesson 1 of 15 · 12 min

# The six-part prompt: role, context, task, constraints, examples, format

## Prompts are briefs

The biggest shift in prompting is to stop thinking of a prompt as a question and start thinking of it as a **creative brief**. If you handed a freelancer the sentence "write a caption for my bakery", you would get something generic. If you gave them your audience, your offer, your tone and the platform, you would get something usable. AI works the same way.

A reliable structure has six parts. You will not always need all six, but checking against them fixes most weak prompts.

## 1. Role

Tell the model what expertise to bring. "You are a senior direct-response copywriter who specializes in Instagram ads for food brands in the Gulf." A role nudges vocabulary, priorities and standards. Keep it relevant; "world's greatest genius" adds nothing.

## 2. Context

This is where most prompts fail. Include:

- **Who you are:** brand, product, price point, positioning.
- **Audience:** who they are, what they want, what they worry about, what language they use.
- **Situation:** campaign goal, platform, season, what has worked before.
- **Facts the model must use:** approved claims, prices, dates.

## 3. Task

One clear verb and deliverable: "Write five caption options", "Rewrite this email", "Summarize these reviews into three themes". If you have several tasks, number them or split them into separate prompts.

## 4. Constraints

The rules: length, tone, words to avoid, compliance requirements, reading level, language variety (British or American English, Gulf or Levantine Arabic, Urdu or Roman Urdu).

Examples: "Under 150 characters before the 'more' cut-off." "No health claims." "Do not use the words 'elevate', 'unlock' or 'game-changer'." "Include #ad at the start because this is a paid partnership."

## 5. Examples

Show, don't just tell. One or two examples of captions you love, or past posts that performed well, teach tone faster than any adjective. (More on this in the next lesson.)

## 6. Output format

Say exactly how you want the answer: a numbered list, a table with columns, a script with timestamps, JSON for a spreadsheet, or "hook on line one, body, then CTA".

## Before and after

**Weak:** "Write a caption for my bakery's new cake."

**Strong:**

> **Role:** You are a social media copywriter for independent food brands in Pakistan.
> **Context:** We are Crumb & Co, a home bakery in Lahore selling custom cakes (from PKR 4,500). Audience: women aged 25–40 planning birthdays, who order via WhatsApp. Our tone is warm, playful and slightly cheeky. We are launching a pistachio-rose cake for Eid.
> **Task:** Write five Instagram caption options for the launch Reel.
> **Constraints:** Each under 60 words. First line must work as a hook on its own. Mix English with light Roman Urdu. No claims about being "the best in Lahore". End with a CTA to WhatsApp.
> **Example of our voice:** "Birthday on Saturday and zero plan? Relax, jaan. We've got the cake. You bring the candles."
> **Format:** Numbered list, each option followed by a one-line note on the angle used.

The second prompt takes two extra minutes to write and saves twenty minutes of editing.

## Where the six parts live in 2026 assistants

You don't have to type all six parts every time. Modern assistants let you store the stable parts once:

| Part | Stable or per-task? | Where to store it |
|---|---|---|
| Role | Mostly stable | Project instructions, a custom GPT, a Gemini Gem, Claude project instructions |
| Context | Brand context stable; campaign context per task | Project files (voice guide, product sheet) + a few lines per task |
| Task | Per task | The prompt |
| Constraints | Brand rules stable; task limits per task | Project instructions + the prompt |
| Examples | Mostly stable | Project files ("best posts", "best emails") |
| Format | Per task, or saved defaults | The prompt, or project instructions for defaults |

With a well-built project, your everyday prompt shrinks to task, task-specific context and format, and the output still reflects all six parts.

## Hands-on: rebuild one of your own prompts

Take the last prompt you used for content or sales. Paste it into this checker:

```text
Here is a prompt I used: [paste].
Check it against six parts: role, context, task, constraints, examples, output format.
For each part: present / missing / weak, and one specific improvement.
Then rewrite the prompt with the improvements, using [PLACEHOLDERS] for facts only I can supply.
Do not invent facts about my business.
```

**Before:** "Write a LinkedIn post about our new HR software feature."

**After:** a structured prompt with a role (B2B SaaS content writer for HR leaders), context (the feature, the audience's pain in their words, one approved customer result as [PROOF]), a single task, constraints (under 180 words, first line under 12 words, no "revolutionary"), one example post, and a format (hook, 3 short paragraphs, question to prompt comments). Fill the placeholders and you'll typically go from four rounds of editing to one.

## Quick checklist

Before you hit enter, ask:

- Would a human freelancer understand exactly what I want?
- Did I give facts the model cannot guess?
- Did I say what to avoid?
- Did I specify the format I need?

## Pitfalls

- **Over-stuffing:** a two-page prompt full of irrelevant history can dilute the key instructions. Put the most important rules near the task, and keep context relevant.
- **Conflicting instructions:** "Be concise" and "include every detail" pull in opposite directions. Decide which wins.
- **Relying on role alone:** "You are an expert marketer" without context still produces generic work.

## Video lecture: The six-part prompt: role, context, task, constraints, examples, format

Lecture coming soon · 11 chapters · about 9 minutes. Read the full transcript below.

1. The six-part prompt
2. Why structure beats cleverness
3. Parts 1 to 3
4. Parts 4 to 6
5. The freelancer analogy
6. Example 1: the Eid cake caption
7. Example 2: the B2B feature launch
8. Watch me do it
9. Store the stable parts
10. Common mistakes
11. Recap and try this now

## Lecture transcript

### The six-part prompt

Imagine handing a freelance copywriter a sticky note that says: write a caption for my bakery. You'd get something generic, and you wouldn't blame the freelancer. You'd blame the brief. AI works exactly the same way. The single biggest shift in prompting is to stop thinking of a prompt as a question and start thinking of it as a creative brief. In this lecture you'll learn a six part structure that fixes most weak prompts, see it applied to a Lahore bakery and a B2B software launch, and watch me rebuild a weak prompt live. You'll also learn where to store the parts so you don't retype them every day.

### Why structure beats cleverness

Why a structure instead of clever tricks? Because most weak output isn't caused by the wrong magic words. It's caused by missing information: who the audience is, what the offer is, what the brand sounds like, what to avoid. A structure makes those gaps visible before you hit enter. You won't always need all six parts. A quick brainstorm might need two. But checking your prompt against the six parts takes ten seconds, and it catches the omission that would otherwise cost you twenty minutes of editing.

### Parts 1 to 3

Let's walk through them. Part one, role. Tell the model what expertise to bring: you're a senior direct response copywriter who specializes in Instagram ads for food brands in the Gulf. A role nudges vocabulary and standards, but world's greatest genius adds nothing. Part two, context, where most prompts fail. Who you are, your product and price point. Your audience: what they want, what they worry about, the words they use. The situation: goal, platform, season, what's worked before. And facts the model must use, like approved claims, prices and dates. Part three, task. One clear verb and deliverable. Write five caption options. If you have several tasks, number them or split them into separate prompts.

### Parts 4 to 6

Part four, constraints. Length, tone, reading level, language variety, words to avoid, and compliance rules. Under sixty words. No health claims. Don't use elevate or unlock. Include hashtag ad because this is a paid partnership. Part five, examples. One or two examples of captions you love teach tone faster than any adjective. The next lesson goes deep on this. And part six, output format. Say exactly how you want the answer: a numbered list, a table with columns, a script with timestamps, or hook on line one, body, then call to action. The right format saves editing time and makes options easy to compare.

### The freelancer analogy

Here's the quick test I use. Before you hit enter, imagine forwarding the prompt to a talented freelancer who has never heard of your brand. Would they understand exactly what you want? Did you give them facts they can't possibly guess, like your price, your offer end date, or your audience's favorite phrase? Did you tell them what to avoid? Did you say what shape the deliverable should take? If the freelancer would email you three questions before starting, the AI will guess the answers to those three questions instead. And its guesses will be average.

### Example 1: the Eid cake caption

Let's see it in action. Weak prompt: write a caption for my bakery's new cake. Now the strong version. Role: social media copywriter for independent food brands in Pakistan. Context: we're Crumb and Co, a home bakery in Lahore selling custom cakes, audience is women planning birthdays who order on WhatsApp, tone is warm, playful and a bit cheeky, launching a pistachio rose cake for Eid. Task: five Instagram caption options for the launch Reel. Constraints: under sixty words, first line works alone, light Roman Urdu, no best in Lahore claims, end with a WhatsApp call to action. Example of our voice: birthday on Saturday and zero plan? Relax, jaan, we've got the cake. Format: numbered, with a note on the angle. Two extra minutes to write. Twenty minutes of editing saved.

### Example 2: the B2B feature launch

Now a B2B example, with illustrative details. A content marketer at an HR software company in London needs a LinkedIn post about a new absence tracking feature. Her old prompt: write a LinkedIn post about our new HR software feature. Her six part version: role, B2B content writer for HR leaders. Context: the feature, the audience's pain in their own words, spreadsheets that are always out of date, and one approved customer result, written as a proof placeholder she fills from the case study. Constraints: under one hundred and eighty words, first line under twelve words, no revolutionary. One example post that performed well. Format: hook, three short paragraphs, a question to prompt comments. Her editing rounds drop from about four to one.

### Watch me do it

Let me rebuild a weak prompt using the checker from the lesson text. Here's the original: write an email to promote our summer camp. I paste it into the checker and ask it to rate each of the six parts and suggest one improvement for each, then rewrite it with placeholders for facts only I can supply, without inventing anything. The table says role is missing, context is missing, task is present but vague, constraints missing, examples missing, format missing. That's harsh, but fair. The rewrite has a role, the audience as parents of eight to twelve year olds, placeholders for dates, price, location and one real parent quote, a two hundred word limit, a subject line and preview text format, and a rule against inventing testimonials. I fill in the placeholders, and I have a prompt I can reuse every summer.

### Store the stable parts

Here's how professionals avoid typing all six parts every day. Split them into stable and per task. The role, the brand context, the brand rules and your best examples rarely change, so store them once: in project instructions and files in ChatGPT or Claude, in a custom GPT, or in a Gemini Gem. The task, the task specific context and the format change every time, so they go in the prompt. With a well built project, your everyday prompt might be three lines, and the output still reflects all six parts. We'll build exactly that kind of context pack later in this course.

### Common mistakes

Three common mistakes. Over stuffing: a two page prompt full of irrelevant history dilutes the key instructions. Keep context relevant, and put the most important rules near the task. Conflicting instructions: be concise and include every detail pull in opposite directions, so outputs flip between them. Decide which wins. And relying on role alone: you're an expert marketer, with no context, still produces generic work. The role sets standards. The context supplies the substance.

### Recap and try this now

Let's recap. A prompt is a brief with six parts: role, context, task, constraints, examples and output format. Context is the part most often missing, so give the facts the model can't guess. Use the freelancer test before you hit enter. And store the stable parts in a project, GPT or Gem so everyday prompts stay short. Here's your try this now. Take the last content or sales prompt you used, run it through the six part checker in the lesson text, fill in the placeholders, and compare the new output with the old one. Save the improved version. It's the first entry in your prompt library.

## Video transcript

Here's the single biggest upgrade you can make to your AI results. Stop writing questions. Start writing briefs.

Think about it. If you told a freelance copywriter, write a caption for my bakery, you'd get something bland. But if you told them who your customers are, what you're launching, how your brand sounds, and what to avoid, you'd get something you could actually post. AI works exactly the same way.

A great prompt has six parts.

One. Role. Tell the model what expertise to bring, like a social media copywriter for food brands in Pakistan.

Two. Context. Who you are, who your audience is, what the campaign is for, and the facts it must use. This is where most prompts fall short.

Three. Task. One clear deliverable. Write five caption options.

Four. Constraints. Length, tone, banned words, and compliance rules, like including hashtag ad on paid partnerships.

Five. Examples. Paste a caption you love. One real example teaches tone faster than a list of adjectives.

Six. Output format. A numbered list, a table, a script with timestamps. Say exactly what you need.

You won't need all six every time. But when an output disappoints, run through this list, and you'll usually find the missing piece.

A strong prompt might take two extra minutes to write. It routinely saves twenty minutes of editing. That's the trade we're going to practice throughout this course.

## Key takeaways

- Treat prompts as creative briefs: role, context, task, constraints, examples and output format.
- Context is the part most often missing; supply facts the model can't guess.
- Store stable parts (role, brand context, rules, examples) in a project, GPT or Gem; type only the task-specific parts.
- Avoid over-stuffing and conflicting instructions; put the most important rules near the task.

## Try it

Rewrite one of your recent weak prompts using all six parts, run both versions, and compare how much editing each output needs.

- [Next: Using examples and output formats to control results](https://optimizeall.com/learn/prompt-engineering-for-content-and-sales/examples-and-output-formats)
- [All lessons of Prompt Engineering for Content & Sales](https://optimizeall.com/learn/prompt-engineering-for-content-and-sales)