AI

Prompt Engineering Best Practices for ChatGPT, With the RTCO Format

Prompt Engineering Best Practices for ChatGPT, With the RTCO Format
Quick answer

Prompt engineering is the practice of designing and refining the text you give ChatGPT so it returns better answers. OpenAI's own help guide boils it down to three habits: be clear and specific, refine your prompt step by step, and describe the tone you want. On top of those, this guide adds the RTCO format: tell ChatGPT its Role, the Task, the Context and the Output you want. RTCO is our own simple checklist, not an official OpenAI term.

If ChatGPT's answers feel vague, too long or off target, the fix is usually in the prompt. OpenAI publishes a short help article, Prompt engineering best practices for ChatGPT, that lists the core ideas. We have gone through it, explained each point with examples, and added the RTCO format so you have one repeatable pattern to copy. It is general education, not financial, legal or medical advice.

What is a prompt?

A prompt is the input you give a large language model (LLM) to start a conversation or trigger a response. Most of the time it is text, such as a question or an instruction. OpenAI's guide notes that a prompt can also take other forms, such as an image or audio.

What is prompt engineering?

Prompt engineering is the process of designing and improving your prompts so the model's response is more useful. You do not need to be technical. In practice it means choosing your words carefully, giving the model what it needs to know, and adjusting when the first answer is not right.

Best practice 1: Be clear and specific

OpenAI's first tip is to make prompts clear and specific, with enough context for the model to understand what you are asking. Avoid ambiguity and be as precise as you can. The model cannot read your mind, so anything you leave out, it guesses.

Vague promptClear and specific prompt
Tell me about index funds.Explain what an index fund is to someone who has never invested, in under 150 words, with one everyday example.
Help me with my email.Rewrite this email to my landlord so it is polite but firm, and ask for a reply by Friday. Keep it under 100 words.
Write a post about AI.Write a LinkedIn post for small business owners about three ways to use ChatGPT to save time on admin. Use plain language and end with a question.

Notice what changed: a clear audience, a clear action, a limit on length, and a detail that shapes the answer. See our longer guide on how to write better prompts for more before-and-after examples.

Best practice 2: Refine your prompt step by step

The second tip is iterative refinement. Your first prompt is a starting point. Read the answer, then adjust the wording, add context, or simplify the request, and try again. Prompt engineering is rarely one-shot.

Useful follow-ups you can send in the same chat:

  • "Make it shorter and keep only the three most important points."
  • "You assumed I am in the US. I am in India, so redo it for that."
  • "Explain the second point again as if I am 12 years old."
  • "Give me three different versions and tell me which one you would pick and why."

Tip: when an answer is close but not right, change one thing at a time. That way you learn which change actually helped.

Best practice 3: Ask for the tone you want

OpenAI's third tip is to describe the tone with adjectives such as formal, informal, friendly, professional, humorous or serious. Their example is "Explain this in a friendly and engaging tone."

Tone changes who the answer feels written for. The same explanation of a budget can sound like a bank letter, a friend's advice, or a lecture, depending on the words you choose. Try pairing two adjectives, for example "warm and professional" or "direct but not rude."

The RTCO format: Role, Task, Context, Output

The three best practices tell you what good prompts look like. RTCO gives you a fill-in-the-blanks way to build one quickly. It is our own checklist, built on the same ideas, and it works for most everyday prompts.

LetterWhat you give ChatGPTExample
RoleThe perspective or expertise to answer fromAct as a patient personal-finance teacher.
TaskThe exact thing you want done, with a clear verbExplain the difference between a stock and a bond.
ContextWho you are, what you know, and any limitsI am a beginner in my 20s, I have never invested, and I only want the basics.
OutputThe shape of the answer: format, length, toneUse a two-column table, then a three-line summary in a friendly tone.

A copy-and-paste RTCO template

RTCO template
Role: Act as [the kind of expert or teacher you want]. Task: [One clear instruction, such as explain, compare, summarise, rewrite or plan]. Context: [Who you are, what you already know, your goal, and any limits]. Output: [Format, length and tone, for example a table, under 200 words, friendly.]

RTCO in action

Example prompt
Role: Act as a patient personal-finance teacher. Task: Explain the difference between an index fund and an ETF. Context: I am a complete beginner and I get confused by jargon. I only want to understand the idea, not to choose a product. Output: A two-column comparison table, then a three-sentence summary in a friendly tone. Under 200 words in total.

Compare that with "Tell me about index funds and ETFs". The RTCO version tells ChatGPT who it is, what to do, who it is talking to and what the answer should look like. You usually get something you can use on the first try.

You do not need all four every time. For a quick question, Task alone can be enough. Add Role, Context and Output when the answer matters or the first attempt missed.

Common mistakes to avoid

  • Asking for too much at once. Break a big request into steps and run them one after another.
  • Leaving out the audience. "Explain this" gives a different answer for a child, a colleague and an expert.
  • Accepting the first answer. Refine. That is the whole point of iteration.
  • Trusting facts without checking. ChatGPT can sound confident and still be wrong. Check important facts, numbers and sources yourself.
  • Sharing private data. Do not paste passwords, ID numbers or confidential documents into a prompt.

Where to go next

OpenAI's guide links to further reading, including its basic prompt engineering guide, its reasoning model prompting guide, and pages on ChatGPT Custom Instructions and ChatGPT capabilities. Custom Instructions are useful because you can save your Role and Context once, so you do not retype them in every chat. You can find all of these through the original OpenAI article.

To practise, try our learn AI prompting through projects plan, or browse the ChatGPT prompts for investors for ready-made examples.

Source: this article summarises the ideas in OpenAI's help article Prompt engineering best practices for ChatGPT in our own words. The RTCO format and the examples are ours. We are not affiliated with OpenAI. ChatGPT is a trademark of OpenAI.

Frequently asked questions

What is prompt engineering?

Prompt engineering is the process of designing and refining the input you give a language model so it responds more usefully. For ChatGPT, that means writing clear prompts and improving them step by step.

What are the best practices for prompting ChatGPT?

OpenAI's help guide lists three general ones: be clear and specific, refine your prompt through iteration, and ask for the tone you want using descriptive adjectives.

What is the RTCO format?

RTCO stands for Role, Task, Context and Output. It is our own checklist: say who ChatGPT should act as, what to do, the background it needs, and the format, length and tone of the answer. It is not an official OpenAI term.

Do I need all four parts of RTCO every time?

No. For a quick question, a clear task can be enough. Add the role, context and output when the answer matters or your first try missed.

Why does ChatGPT give different answers to the same prompt?

It generates responses rather than looking up a fixed answer, so wording changes and chance can produce different results. Adding detail and refining the prompt makes answers more consistent.

Can I save my role and context so I do not retype them?

OpenAI's guide points to ChatGPT Custom Instructions, which let you store preferences so they apply to your chats. Check OpenAI's help page for how it works in your plan.

Is this official OpenAI guidance?

The three best practices summarise OpenAI's public help article. The RTCO format and examples are ours. We are not affiliated with OpenAI.

Penny the squirrel holding an acorn

Take your first step

Ready to start using AI, one small step at a time?

Join the free live AI Masterclass and see how to start using AI in your work, step by step. No experience needed.

Join the free AI Masterclass → Want hands-on practice first? See the AI Breakthrough