---
title: "Prompting fundamentals that work in ChatGPT"
description: "The R-G-C-F structure For any prompt that matters, give ChatGPT four things: - Role / context: who you are, who the audience is, the situation. (\"I'm the…"
url: https://optimizeall.com/learn/mastering-chatgpt/prompting-fundamentals-in-chatgpt
updated: 2026-10-05
---

Mastering ChatGPT (OpenAI) · ChatGPT foundations · lesson 2 of 19 · 14 min

# Prompting fundamentals that work in ChatGPT

## The R-G-C-F structure

For any prompt that matters, give ChatGPT four things:

- **Role / context:** who you are, who the audience is, the situation. ("I'm the marketing lead at a UK online bakery; this goes to wholesale buyers.")
- **Goal:** the one job you want done, stated plainly.
- **Constraints:** length, tone, spelling, what to avoid, facts that must not change, compliance rules.
- **Format:** the output shape: table with named columns, email with subject line, bullets, JSON.

Current GPT models follow instructions closely, so be explicit. If you want options, ask for options. If you want no preamble, say "no preamble". If a rule matters, say why it matters.

## Delimiters and examples

Separate pasted materials with clear delimiters and labels:

```text
### Brief
Launch of our Eid gift hampers, UAE and KSA, three price points.

### Brand rules
Warm, respectful tone. No discount language in week one. English and Arabic.

### Approved past posts
1) ...
2) ...

Task: Write 5 Instagram captions following the brand rules, in the style of
the approved posts. Return a table: Caption (EN) | Caption (AR) | CTA | Rule check.
```

**Few-shot prompting** (showing one or two examples of the output you want) is the strongest steering tool. Say what to copy (structure, tone, length) and what not to copy (specific facts).

## Ask it to ask

End important prompts with: *"Ask me up to three questions before you start if anything important is missing."* This reduces guessing and invented details more than any clever phrasing.

## Iterate precisely

- **Edit your original message** when you realise you left out context; the conversation branches from the improved prompt.
- **Give editor-style feedback** when the draft is close: "Keep the structure. Cut to 120 words. Replace the second paragraph with a customer example."
- **Regenerate** only when the prompt was good and you want a genuinely different alternative.
- **Start a new chat** when a long conversation has drifted, with a short summary of decisions so far.

## Formats that save time downstream

Ask for outputs you can paste straight into other tools:

- "A table I can paste into Google Sheets: Date | Platform | Hook | Caption | CTA."
- "A CSV with headers name,email_subject,body."
- "Markdown with H3 headings for our CMS."
- "A checklist with checkboxes."

## Worked example: from vague to usable

**Vague:** "Write an email to wholesale customers about our new range."

**R-G-C-F:**

```text
Role/context: I run a small artisan bakery in Leeds that supplies 40 cafés.
Audience: café owners who reorder weekly; they read email on their phones.
Goal: Announce our new gluten-free range and get them to add one item to next week's order.
Constraints: 120-150 words; British spelling; friendly, not salesy; don't claim
"coeliac-safe" (we're not certified); include order cut-off "Thursday 2pm".
Format: subject line (under 45 characters), preview text, body, one-line P.S.
Ask me up to 3 questions first if anything important is missing.
```

ChatGPT asked for the product names and prices, then produced a draft that needed one edit. Note the compliance constraint: it prevents a claim the bakery cannot substantiate.

## Hands-on: your prompt library

1. Create a document called *Prompt library*.
2. Rewrite your three most-used prompts with R-G-C-F, delimiters and one example each.
3. Run old and new versions and compare the edit time.
4. Save the winners. You will reuse them in Projects and custom GPTs.

## Breaking big jobs into stages

For anything longer than a page, split the work into stages you approve one at a time:

1. **Outline:** "Propose three angles and an outline. Don't draft yet."
2. **Draft:** "Draft section 1 only, using the approved outline."
3. **Critique:** "Review as a sceptical buyer; list the five biggest weaknesses; don't rewrite."
4. **Revise:** "Fix weaknesses 1, 2 and 4. Keep everything else."
5. **Package:** "Give me a subject line, a 40-word summary and a table version for Sheets."

Each stage is small enough to check properly, and a mistake in the outline costs seconds rather than a full rewrite. Save the stage prompts together in your library as a reusable "long document" recipe.

## Reusable prompt template

```text
Role/context: [who I am, audience, situation, why it matters]
Goal: [one job]
Constraints: [length, tone, spelling, must-not-change facts, compliance rules]
Format: [exact output shape]
Examples: [1-2 labelled examples, with what to copy]
Ask me up to 3 questions first if anything important is missing.
```

## Pitfalls

- Contradictory constraints ("comprehensive, under 40 words").
- One giant prompt for a multi-stage job; break it into outline, draft, critique, revise.
- Negative-only instructions; say what to do instead.
- Pasting passwords, card data or confidential client information.

## How to measure success

More first drafts you can use with light edits, fewer back-and-forth turns, and a growing library of prompts your team reuses.

## Video lecture: Prompting fundamentals that work in ChatGPT

Lecture coming soon · 16 chapters · about 9 minutes. Read the full transcript below.

1. Prompting that works
2. Why structure matters
3. R-G-C-F
4. Delimiters keep materials apart
5. Examples beat adjectives
6. Ask it to ask
7. Iterate precisely
8. Formats for downstream tools
9. Simple example
10. Worked example: a Leeds bakery
11. Big jobs in stages
12. Pitfalls
13. Try this now
14. Watch me do it, part 1
15. Watch me do it, part 2
16. Recap and next step

## Lecture transcript

### Prompting that works

There are no magic words in ChatGPT. There is, however, a reliable structure that turns vague requests into usable first drafts. In this lecture you will learn the four part R G C F structure, how to separate pasted material, why examples beat adjectives, and how to iterate like an editor.

### Why structure matters

Why bother with structure? Because whatever you leave out, the model fills in with an average guess. No audience means a generic audience. No length means whatever length feels typical. No tone means safe and bland. Imagine briefing a designer with just the words, make a poster. You would get a poster, but not yours. The four part structure is simply a checklist that makes sure your specific details replace the model's average ones.

### R-G-C-F

R stands for role and context. Who you are, who the audience is, and the situation. G is the goal, one job stated plainly. C is constraints, meaning length, tone, spelling, things to avoid, facts that must not change and compliance rules. F is format, the shape of the output, a table with named columns, an email with a subject line, or bullet points. Current GPT models follow instructions closely, so be explicit. If a rule matters, say why.

### Delimiters keep materials apart

When you paste several things, a brief, brand rules and past posts, separate them with clear labelled headings or delimiters. It stops details bleeding from one document into another, and lets you refer to each precisely. Check every caption against the brand rules section.

### Examples beat adjectives

Few shot prompting, which just means showing one or two examples of what you want, is the strongest steering tool you have. It beats a paragraph of adjectives like punchy and on brand. Say exactly what to copy, such as the structure, tone and length, and what not to copy, such as specific product names or prices.

### Ask it to ask

Add one line to important prompts. Ask me up to three questions before you start if anything important is missing. Now, instead of inventing your price or cut off time, ChatGPT asks. This single habit reduces invented details more than any clever wording.

### Iterate precisely

Then iterate precisely. If you left out context, edit your original message so the conversation branches from the better prompt. If the draft is close, give editor style feedback, keep the structure, cut to one hundred and twenty words, replace paragraph two with a customer example. Regenerate only when the prompt was good and you want a different alternative. And when a long chat drifts, start fresh with a short summary.

### Formats for downstream tools

Ask for formats that save you time afterwards. A table you can paste straight into Google Sheets, with named columns. A CSV with specific headers for your email tool. Markdown with H three headings for your website. A checklist with boxes. The less reformatting you do, the more time AI actually saves.

### Simple example

A simple example. Write a product description for our dates, gives you something that could describe any dates on earth. Now the structured version. Role and context, I sell premium Medjool dates online in the UK, customers are gift buyers. Goal, a description for the product page. Constraints, sixty words, British spelling, no health claims. Format, one short paragraph then three bullet features. The second version is ready to publish with a light edit, because every choice that mattered came from you.

### Worked example: a Leeds bakery

A small bakery in Leeds supplies forty cafés and wants to announce a gluten free range. The vague prompt produced a generic email. The structured version set the audience, café owners reading on their phones, the goal, add one item to next week's order, a length of one hundred and twenty to one hundred and fifty words, and one vital constraint, do not claim coeliac safe because the bakery is not certified. ChatGPT asked for product names and prices, then delivered a draft needing one small edit, and no claim the bakery could not back up. The bakery then reused the same prompt for a second announcement, a seasonal hot cross bun range, by changing only the goal and the product details. The first draft needed no edits at all, because the audience, tone, length and compliance rule were already baked into the template. That is the real payoff of structure. The second time is almost free.

### Big jobs in stages

For anything longer than a page, work in stages you approve one at a time. First the outline, with three angles and no drafting yet. Then one section at a time. Then a critique from a named reader, without rewriting. Then targeted revisions, fix weaknesses one, two and four, keep everything else. Finally, package it, a subject line, a short summary and a table version. A mistake caught at the outline stage costs seconds. The same mistake caught in a finished draft costs an hour.

### Pitfalls

Four pitfalls to avoid. Contradictory constraints, like comprehensive but under forty words. Giant prompts for multi stage jobs. Negative only instructions like do not be boring, instead of saying what you want. And pasting passwords, card data or confidential client details. Measure success by the share of first drafts you can use with light edits.

### Try this now

Try this now. Open a document called prompt library. Find the three prompts you use most often and rewrite each one with role, goal, constraints and format, plus one short example of what good output looks like, and the line, ask me up to three questions first. Run the old and new versions side by side and note how much editing each needs. Save the winners. Over the next month, every time a prompt works really well, add it. That library becomes one of your most valuable work assets.

### Watch me do it, part 1

Let me build the Leeds bakery email live. Role and context. I run a small artisan bakery in Leeds supplying forty cafés, the readers are café owners reading on their phones. Goal. Announce our new gluten free range and get them to add one item to next week's order. Constraints. One hundred and twenty to one hundred and fifty words, British spelling, friendly not salesy, do not claim coeliac safe because we are not certified, include the order cut off, Thursday at two. Format. A subject line under forty five characters, preview text, the body and a one line P S. Last line, ask me up to three questions first if anything important is missing.

### Watch me do it, part 2

ChatGPT asks for the product names and prices, and I answer in one line. The draft arrives at one hundred and forty two words. The subject line is thirty eight characters, and there is no coeliac claim, just made without gluten containing ingredients, which our baker confirms is accurate. I give editor feedback. Keep the structure, shorten the P S to one sentence. Then I ask for three subject line variants as a table I can paste into Google Sheets, with columns for subject, preview text and angle. Finally I save the full prompt to my prompt library under wholesale announcements.

### Recap and next step

Recap. Use role, goal, constraints and format. Separate materials with delimiters, steer with examples, invite questions, and iterate precisely. Your next step is to start a prompt library, rewrite your three most used prompts with this structure, and compare how much editing the results need. You will reuse the winners in Projects and custom GPTs.

## Key takeaways

- Use Role/context, Goal, Constraints and Format for important prompts.
- Separate pasted materials with delimiters and include examples of good output.
- Edit the original prompt to fix gaps; regenerate only for fresh alternatives.
- Ask for structured formats such as tables that you can paste into other tools.

## Try it

Rewrite three prompts you used this month using R-G-C-F with one example each. Compare the before and after outputs.

- [Previous: Models in 2026: fast answers vs thinking, and how to choose](https://optimizeall.com/learn/mastering-chatgpt/reasoning-vs-fast-models)
- [Next: A guided tour of the ChatGPT workspace](https://optimizeall.com/learn/mastering-chatgpt/chatgpt-workspace-tour)
- [All lessons of Mastering ChatGPT (OpenAI)](https://optimizeall.com/learn/mastering-chatgpt)