The RTCF formula in practice

You already know the RTCF formula — Role, Task, Context, Format — from the free track. Now let's see how each part actually changes the answer.

Role sets the expertise and the tone

Compare two answers to the same question, "how do I raise my prices":

• No role — generic advice you could have googled.
• "You're a B2B pricing consultant" — frameworks, segmentation, churn risk.

A role isn't magic. It's a filter: it tells the model which part of what it knows to answer from.

Task: one verb, one result

Weak tasks glue several actions together: "take a look, fix it up and make it better." Strong ones name a single action and a result: "Find 5 factual errors", "Cut this to 100 words, keep every number".

Context is everything the model can't know

It doesn't know your company, your audience or your constraints. Put them in the prompt: "We sell CRM software to small businesses, average deal is €150/month, our main competitor keeps undercutting us". The rule: if you were handing this task to a new hire, what would they ask you? Those answers are your context.

Format saves you the rewrite

The skill someone was paid $335,000 for

In early 2023, Anthropic (the company behind Claude) posted a "prompt engineer" job ad paying up to $335,000 a year — and the headlines went off: "a career with no degree required", "AI whisperers". Sceptics rolled their eyes: "it's just typing questions." Three years on, the truth turned out to be more interesting than either take. The standalone job title mostly dissolved — but the skill itself became table stakes for anyone who works with their head.

"Can brief an AI" now shows up in job ads for marketers, lawyers, analysts and managers, the way "comfortable with computers" once did. You're not learning a party trick. You're learning the new literacy.

One prompt, four versions

Watch a request turn from junk into a tool. The job: a price-increase message to the members of a fitness studio.

v1, bare request: "write something about the price increase" → a faceless brush-off that will annoy every member who reads it.

v2, + role: "You own a fitness studio and you genuinely care about your members. Write…" → warmer, but still about prices in the abstract.

v3, + context: "…Prices haven't moved in 3 years, our rent went up. A 12% increase from March 1. Long-standing members get a 2-month freeze" → now there are actual reasons and specifics, and the message reads honest.

v4, + format: "…Structure: thank them → the reason in one sentence → what changes → what they get in return → a warm close. Under 120 words, never use the word 'unfortunately'" → ready to send, no edits.

One minute of work separates v1 from v4. The gap in the result is the gap between an intern and someone who's done this for years. Before you read on, be honest: which version do you usually stop at?

Quick FAQ

• "Do I always need all four?" No. For "what's the capital of Australia" the formula is overkill — it's for work you'll actually use: emails, copy, plans, analysis. Rule of thumb: the more a mistake costs you, the fuller the prompt.
• "Does the order matter?" Not much, but the classic order reads better: role → task → context → format. The model will cope with a jumble; you, re-reading your own prompt a month later, might not.
• "Do I have to type all this every time?!" No — in the templates lesson you'll learn to save the skeleton and drop it in within seconds, and the constant stuff (who you are, how you like answers) moves into custom instructions, the permanent assistant settings covered in part 2 of the free track. The formula is a way of thinking about the task, not a typing ritual.

One last thing: the formula travels. RTCF works the same in ChatGPT, Claude and Gemini, on free plans and paid ones. You're investing in a skill that will outlive every model and pricing change to come.

Do this now

Take the last thing you asked an AI and run it through the four parts: is there a role? is the task one action? would a new hire have enough context? did you name a format? Fill in what's missing and compare the two answers. Five minutes of work, and the penny usually drops for good.