0.4Prompting Copilot well
The essentials in 30 seconds
- The RTCF structure: Role (“As a…”) + Task (“write…”) + Context (“for whom, why”) + Format (“length, table…”) — format is the most profitable and most forgotten ingredient.
- Iterate rather than retype: refine through successive exchanges (“shorter”, “as a table”) — the conversation keeps the context, you build step by step.
- One example beats ten explanations: provide a model of the desired result (“following this model: …”) — the most effective technique for getting THE right format.
Threaded mini-case — Hicham writes his remote-work memo in Algiers
Hicham, administrative assistant in Algiers, must announce 2 remote-work days per week. First lazy attempt: “Write a memo on remote work” — generic text to rewrite entirely. He applies the RTCF method: role (HR assistant), task (service memo), context (all employees, clear and reassuring tone), format (one page, 3 sections, bullets for terms). The result is close to the final deliverable; two iterations (“shorter”, “as a table for the terms”) and one in-house header example are enough. He archives this prompt: the first item in his library.
Standalone case for this section: it returns in the lesson, the diagram, the script and the 48h challenge of section 0.4.
1. The RTCF structure
A prompt is your instruction to the AI: the more precise and contextualised it is, the better the answer. The professional recipe has four ingredients — the RTCF structure: Role (“As an HR manager…”), Task (“write a memo…”), Context (“for new joiners, caring tone…”), Format (“one page, with bullets”). The output format is most often forgotten — and that is where time is won. ALWAYS specify length, structure, whether a table is included: otherwise you get the “standard essay” to rework entirely. Hicham checks his prompts on his fingers: R, T, C, F — are all four there?
2. Iterate rather than retype
Iteration is the reflex that distinguishes advanced users: refining through successive exchanges (“shorter”, “more formal”, “as a table”) rather than reformulating everything from scratch. The conversation keeps the context: you build the answer step by step. You can also exploit the model's variability to your advantage — its implicit temperature: asking for “3 proposals in different tones” gives you a choice, instead of a single take-it-or-leave-it answer. Hicham never opens a new conversation for an adjustment: he continues the thread.
3. One example beats ten explanations
The most effective technique for getting THE right format is called the example (few-shot): providing a model of the expected result (“following this model: …”). Copilot then mirrors the structure, tone and length — without you having to describe them in ten lines. There is no need, by contrast, to be “polite” (“please”, “thank you”): no measurable impact on quality. Write naturally; what counts is the precision of the task, context and format. Hicham always attaches an in-house header when he wants an exact format.
🔎 Concrete example — 🔎 Concrete example — Vague prompt vs RTCF prompt
Hicham wants a service memo on remote work. Compare two wordings.
Vague: “Write a memo on remote work.”
RTCF: “As an HR assistant (role), write a service memo announcing 2 remote-work days per week (task), for all employees, clear and reassuring tone (context). Format: one page, with header, 3 sections (principle, terms, contact) and a bullet list for the terms (format).”
Result: the RTCF prompt gives an almost final deliverable; the vague prompt, generic text to rewrite.Reading: same need, same AI — only the structure changes: RTCF frames the role, task, context and above all the format.
Lesson: each RTCF line saves a rework pass: specifying costs less than trimming.
Section vocabulary
- Prompt
- Your instruction to the AI. The more precise and contextualised it is, the better the answer.
Simple example: Hicham replaces “Write a memo” with a full RTCF prompt.
What it is not: a simple question asked at random with no framework.
Reflex: before sending, check R, T, C, F.
- RTCF structure
- Role + Task + Context + Format: the 4 ingredients of the professional prompt.
Simple example: Hicham's remote-work memo ticks all four boxes.
What it is not: a magic formula that would guarantee a perfect result.
Reflex: use it for every serious request.
- Iteration
- Refining through successive exchanges (“shorter”, “more formal”, “as a table”) rather than reformulating from scratch.
Simple example: Hicham chains “shorter” then “as a table” in the same thread.
What it is not: starting over in a new conversation with each attempt.
Reflex: continue the thread while adjusting the same deliverable.
- Example (few-shot)
- Providing a model of the expected result (“following this model: …”): the most effective technique for getting THE right format.
Simple example: Hicham pastes his in-house header and Copilot mirrors it at once.
What it is not: a long abstract description of the desired format.
Reflex: when the format resists, show instead of describing.
- Implicit temperature
- Asking for “3 proposals in different tones” exploits the model's variability to your advantage.
Simple example: Hicham picks the best of the three tones proposed for his memo.
What it is not: a technical setting reserved for developers.
Reflex: ask for variants when hesitating on tone.
In practice — In practice — Write an RTCF prompt on a real task
- Pick a recurring task (a standard email, a summary, a memo). First write the request “as usual”, like Hicham's first attempt.
- Rewrite it in RTCF format: explicitly add the Role, Task, Context and above all the output Format.
- Run both versions in Copilot and compare: the RTCF version should approach the final deliverable.
- Refine by iteration (“shorter”, “as a table”) and provide an example if the format is not exactly right.
Ready-to-say script — teaching RTCF in one minute (45 s)
“A good prompt has four ingredients: Role — who I speak as; Task — what to do; Context — for whom, in what tone; Format — length, structure, table or bullets. Format is where time is won. Then iterate in the same thread — shorter, more formal — instead of retyping everything. And when the format resists, show an example. Like Hicham, I keep my prompts that work in a library.”
Say while writing R-T-C-F on the board; have them recite the four letters at the end.
Copy-paste template — my RTCF service memo
As an administrative assistant (role), write a service memo announcing 2 remote-work days per week, Tuesdays and Thursdays, from March (task), for all employees at our Algiers site, clear and reassuring tone (context). Format: one page maximum, with header, 3 sections (principle, terms, contact) and a bullet list for the terms (format).
Iteration 1: Half as long, without changing the information.
Iteration 2: Present the terms as a table (day / place / rule).
Iteration 3: Propose 3 different email subjects to share this memo.Standard email
Hicham adapts the template to a recurring email: role (department assistant), task (acknowledgement), context (external requester, courteous tone), format (subject + 80 words + signature) — archived in his library.
Meeting report
After a meeting in Algiers: task (meeting report), context (participants, expected decisions), format (10 bullets + actions/owners/deadlines table) — fed with his rough notes.
Free version
Without a 365 licence: same RTCF structure on neutral content (definitions, plans, examples) — the method trains without internal data.
Demanding pair
Hicham and his colleague review each other: one checks that the four letters R-T-C-F are present, the other that the format is measurable (length, structure) — otherwise, reformulate.
✓ Best practices
- Systematically specify the output format: it is the most profitable and most often forgotten ingredient.
- Build YOUR library of prompts that work for your recurring tasks — the purpose of the thread project.
- Paste an example of the desired result when the format is hard to describe: Copilot mirrors it immediately.
✗ Common pitfalls
- Forgetting the output format: you get the “standard essay” to rework entirely.
- Reformulating everything from scratch with each attempt instead of iterating: you lose the conversation context.
- Believing that politeness (“please”) improves the answer: only precision counts.
Key takeaways
- Role + Task + Context + Format: the structure to use for every serious request.
- ALWAYS specify the output format (length, structure, table…): that is where time is won.
- Iterate instead of retyping: the conversation keeps the context.
- One example of the desired result beats ten lines of explanation.
48h challenge — apply tomorrow
Take a recurring task like Hicham, run the vague then the RTCF version of the template, compare — then archive your first library prompt.
- I can name the four RTCF letters and their content.
- I compared vague vs RTCF on a real task and measured the gap.
- I archived my first reusable prompt in my library.
Frequently asked questions
Should I be polite with Copilot (“please”, “thank you”)?
No measurable impact on quality — write naturally. What counts: precision of task, context and format.
Are there libraries of ready-made prompts?
Yes: Microsoft offers “Copilot Prompt Gallery” (in the Copilot app and on the web), with prompts by profession to adapt. Above all, build YOUR library of prompts that work for your recurring tasks — the purpose of the thread project.
Hicham gets generic text to rewrite. What did he most likely forget?