Prompting properly with RCICF

Listen instead (audio briefing)

A five-part structure that turns vague requests into answers you can use first time, with any AI tool.

Why most prompts disappoint

Most poor AI answers are not caused by a poor tool. They are caused by an incomplete request. When you type “write an email about the late delivery”, the AI doesn’t know who you are, who the email is for, what has happened, or what a good result looks like. So it fills every gap with something safe, generic and bland.

A useful test: imagine giving your request to a capable new colleague on their first day. Would they know exactly what to do, or would they come back with five questions? If they’d need to ask, the AI needs that information too. It just won’t ask; it will guess.

The fix is not longer prompts. It’s complete ones. RCICF is a simple checklist of the five things a complete request contains: Role, Context, Instruction, Constraints and Format.

Role: who should the AI be?

The role tells the AI which expertise, vocabulary and point of view to use. “You are an experienced credit controller” produces a different answer from “you are a friendly customer service agent”, even for the same task.

Choose a real, specific role that matches the job: a payroll specialist, a plain-language editor, a sceptical procurement manager, a coach for new team leaders. You can also give the AI a role as your critic: “act as a demanding client reviewing this proposal”.

Keep it realistic. Fantasy roles such as “the world’s greatest writer” add nothing useful.

Context: what does it need to know?

Context is the background a colleague would need: the situation, who the audience is, why the task matters, and any material to work from. It is usually the part people leave out, and it makes the biggest difference.

Good context answers four questions. What’s going on? Who is this for? What’s the purpose? What information should it use? If you have source material such as notes, an email thread or figures, paste the relevant part and say so: “use only the figures below”.

Leave out personal information the AI doesn’t need, such as client names or ID numbers. Module 6 covers this in detail.

Tip: Short on time? Context is the one part never to skip. A prompt with only an instruction and good context usually beats a prompt with all the other parts and no context.

Instruction: what exactly should it do?

The instruction is the task itself. Start with one clear verb: write, summarise, compare, list, rewrite, explain, draft, check.

Ask for one main thing per prompt. If the job is big, break it into steps: first ask for an outline, agree it, then ask for each section. You get far better results from three focused requests than from one enormous one.

Constraints: what are the limits?

Constraints set the boundaries: length, tone, reading level, what must be included and what must be avoided. For example: “under 150 words”, “warm but professional”, “don’t mention pricing”, “use South African English”.

One constraint belongs in almost every work prompt: “Don’t invent facts, figures or references. If something is missing, say so.” It greatly reduces made-up content.

You can also ask the AI to check with you first: “Before you answer, ask me up to three questions if anything is unclear.”

Format: what shape should the answer take?

Format describes how the answer should look, so you can use it straight away. A table with named columns. Five bullet points. An email with a subject line and three short paragraphs. A one-page brief with headings.

If you have an example of what good looks like, paste it and say “use the same structure as this example”. Showing beats describing.

Putting it together

Here is a one-line request turned into a complete prompt. Each coloured row is one RCICF part. Notice that none of it is clever wording: it’s simply the information a colleague would need.

A complete prompt split into its five RCICF parts: role, context, instruction, constraints and format
The same request, complete. Each part answers a question the AI would otherwise guess.

Worked example: from vague to usable

Thandi coordinates operations for a transport company. She has to tell twelve site managers about a new fuel-card rule. Here are her two attempts.

Worked example: A new fuel-card rule for site managers

Before: Prompt

Write a message about the new fuel card rule.

What came back

Dear Team, We are pleased to inform you of an important update to our fuel card policy. Going forward, all employees are expected to comply with the new guidelines. Please ensure you familiarise yourself with the policy and reach out with any questions. Thank you for your cooperation.

Polite and useless: no date, no rule, nothing to do, and the AI couldn’t have known any of it.

After: Prompt

You are an operations coordinator writing to 12 busy site managers who read messages on their phones. From 1 December, fuel cards can only be used at our two contracted suppliers (the list is in the fleet app), and every slip must be photographed in the app within 24 hours. A card used anywhere else gets one warning, then it’s blocked. Write a WhatsApp message announcing the change. Under 90 words, friendly but clear, no jargon. Don’t invent details. Start with the date and the rule, then what to do, then who to ask: Sipho, extension 214.

What came back

Fuel cards: new rule from 1 December. Cards will only work at our two contracted suppliers (list in the fleet app). After every fill-up, photograph the slip in the app within 24 hours. A card used anywhere else gets one warning, then it’s blocked. Please share this with your drivers this week. Questions? Sipho, ext. 214.

Ready to send after one check. Every fact in it came from the prompt.

Before sending, Thandi checked the start date with finance and confirmed the supplier list was in the app. The AI shaped the message; she supplied and checked the facts.

Five mistakes that weaken prompts

Asking for everything at once. “Write the whole tender response” produces a long, shallow draft. Ask for an outline, agree it, then build it section by section.

Hiding the audience. The same content for a board member, a new hire and a client needs three different answers. Always say who will read it.

Forgetting the material. If the answer depends on your figures, notes or policy, paste them. Otherwise the AI uses general knowledge, which may be wrong for your business.

Accepting the first draft. The first answer is a starting point. One round of specific feedback usually takes it from 70% to 95%.

Retyping good prompts. If you’ve written a prompt that works, save it. Rewriting it from memory each week wastes the effort you already put in.

Improve, don’t restart, and save what works

Treat the first answer as a draft, not a verdict. Tell the AI exactly what to change: “shorter”, “warmer in the opening”, “add the order number”, “make the second point a table”. Specific feedback gets specific improvements.

When a prompt works well, save it in a personal prompt library: a simple document or spreadsheet with a name for each prompt, the prompt itself, and the tool you used. Next time, copy, change the details and go. Over a few weeks, this library becomes one of your most valuable work tools.

Four steps: first draft, specific feedback, better draft, save the prompt, with a loop back for another round of feedback
Each round of specific feedback improves the draft. When it’s right, save the prompt, not just the answer.

Key points

  • A complete prompt has five parts: Role, Context, Instruction, Constraints and Format.
  • Context is the part people skip, and it makes the biggest difference.
  • Always add: don’t invent facts; if something is missing, say so.
  • Improve the first answer with specific feedback, and save prompts that work.

Exercise: Rewrite one prompt you really use

Your version: pick the line for your area of work
  • Operations: a delay or exception notice you send to customers in your industry
  • Finance: explaining a budget variance to a non-finance manager
  • Sales and marketing: a LinkedIn post announcing a new product or service
  • HR and people: an invitation to a performance review conversation
  • Executive team: an update to your board on a project that is behind schedule
  • Other (any role): a message you send often in your work as your role

You’ll take a request you make often, your task (see “Your version” above), turn it into a complete RCICF prompt, improve it once and save it. Allow 20 minutes. Use your company’s approved AI tool.

Step 1: Run your usual prompt

Open your approved AI tool. Type the request for your task (see “Your version” above) the way you normally would, in one line. Copy the answer into a note: you’ll compare it later.

  • ☐ You used your company’s approved tool
  • ☐ No client names, ID numbers or account details in the prompt

Step 2: Fill in the five parts

Copy this template into a note and replace every [bracket] with real details from your work. Write it the way you’d brief a capable new colleague.

Prompt to copy:

You are [a specific role, for example an experienced your area specialist in your industry].

[Context: what's going on, who will read the answer, why it matters. Paste the notes, figures or email it should use.]

[Instruction: one clear verb and the task, for example “Write…”, “Summarise…”, “Compare…”.]

[Constraints: length, tone, must include, must avoid.] Don't invent facts or figures. If something is missing, say so.

[Format: for example a subject line and three short paragraphs, or a table with named columns.]

Example: See the fuel-card example in the lesson for a filled-in version. In your industry, a good one to try is the example for your industry in the list below.

Step 3: Run it and compare

Run the new prompt in a new chat. Put the two answers side by side and score each one out of 5 on three things.

  • ☐ Accurate: no invented facts, the details are right
  • ☐ Right for the reader: tone, length and level fit the audience
  • ☐ Ready to use: you could send or file it with light edits

Step 4: Give one round of specific feedback

Tell the AI exactly what to change, in one message. Name up to three changes. Avoid “make it better”.

Prompt to copy:

Keep everything else the same, but:
1. [first change, for example “make the opening warmer”]
2. [second change, for example “add the order number in the first line”]
3. [third change, for example “cut the last paragraph”]

Example: Keep everything else the same, but: 1. lead with the new date, 2. replace “we apologise for any inconvenience” with one specific sentence of apology, 3. end with the name of the person to contact.

Step 5: Save it

Add the final prompt to your prompt library: a name, the prompt with [brackets] for the parts that change, the tool, and one line on which RCICF part made the biggest difference. The Prompt library template in Insights has the columns ready.

Hand in: Paste your final prompt here. It’s saved with your notes, so you can copy it again later.

Examples by industry: show the learner the line for their sector, or list them all
  • Finance & Banking: a reply to an SME client whose overdraft review has been delayed for a second week.
  • Retail: a reply to a customer whose couch arrived damaged and two weeks late.
  • Insurance: a reply to a policyholder whose storm claim has been waiting for an assessor for ten days.
  • Medical & Healthcare: a reply to a patient whose medical aid has rejected a claim for their consultation.
  • Sports: a reply to a parent unhappy that their child wasn’t selected for the first team.
  • Travel & Tourism: a reply to a corporate traveller whose flight was cancelled the night before a client meeting.
  • Legal: a reply to a client asking why their matter has taken three months longer than quoted.
  • IT & Technology: a reply to a manager whose team has been without email for half a day.
  • Human Resources: a reply to an employee whose leave application was declined during a busy period.
  • Logistics: a reply to a client whose urgent load is stuck at the border.
  • Manufacturing: a reply to a customer whose order will ship a week late because of a material shortage.
  • Public sector and education: a reply to a resident whose complaint about a burst pipe has had no response.
  • Agriculture: a reply to an exporter whose container of citrus was rejected at the port.
  • Mining: a reply to a community leader complaining about dust and blasting times.
  • Energy & Utilities: a reply to a business customer asking why their solar installation is late.
  • Property & Construction: a reply to a client unhappy about a delay in the handover of their unit.
  • Media & Publishing: a reply to a reader complaining that an article misquoted them.
  • Telecommunications: a reply to a business customer whose fibre has been down for two days.

You’re done when: You have a saved RCICF prompt, a before-and-after comparison, and you know which part improved the answer most.