A short, hands-on workshop that teaches prompts as clear briefs: context, objective, constraints, examples, and evaluation. About 25 minutes on screen — and the print version is the participant worksheet.
Workshop · Sine Language
ContextObjectiveConstraintsExamplesEvaluation≈ 25 min
Forget magic words. A prompt is not an incantation — it is a brief, and you already know how to write one. You do it every time you hand work to a colleague.
Step 01 · The problem
The ask that wastes an afternoon.
Most disappointing output starts as a reasonable-sounding request. Here is one, and here is everything it quietly left to chance.
The request, as typed10 words of instruction — 0 of situation
Can you summarize the attached document for the team?
Every unstated decision still gets made. The tool makes it — fluently, confidently, and without telling you which parts it guessed. A vague ask does not save time; it moves the time into the revision cycle, where it costs more.
Guess 1
For whom? Claims processors, leadership, and new hires need different summaries.
Guess 2
To do what? Reference, decision, action — the purpose changes what survives the cut.
Guess 3
How long? Three lines or three pages are both "a summary."
Guess 4
In what shape? Email, table, briefing note, talking points.
Guess 5
What stays out? Names, numbers, caveats, half-decided items.
Guess 6
Checked how? Nothing says what must be verified before anyone relies on it.
Step 02 · The structure
The five-part brief.
One structure, five parts, in the order you would brief a person. It is not the one perfect prompt formula — there is no such thing — but it makes every silent guess visible, and it transfers across tools. If your workplace teaches GCSE, RCCO, or another acronym: same bones, different labels.
Part 01
Context
What can the assistant not see?
Definition
The situation the assistant has no access to: who you are, who the work is for, which documents matter, and what is happening around the task.
Example
# CONTEXT You are helping a team lead at a B.C. benefits carrier. The attached update changes how physiotherapy claims are assessed from 1 September. The team applies these rules daily.
Why it matters
The assistant starts from zero. Anything you do not state, it fills with the most statistically ordinary assumption — and your situation is rarely the ordinary one.
Part 02
Objective
What does done look like?
Definition
The deliverable, named precisely, with a definition of done. One task per brief — a second task gets a second brief.
Example
# OBJECTIVE Produce a one-page summary the team can act on: what changed, what stays the same, what to do differently from 1 September. Done means a claims processor could apply the new rules after one read.
Why it matters
"Summarize this" names an activity. An objective names a result that someone could check. The difference is guess 2, guess 3, and half of guess 5.
Part 03
Constraints
What rules must the work respect?
Definition
The boundaries: length, format, tone, what to exclude, and what to do when the source is unclear. A constraint you can test beats a quality you can only admire.
Example
# CONSTRAINTS One page. Plain language, no legal phrasing. No member details or personal information. Where the update is ambiguous, flag it — do not guess.
Why it matters
Constraints are also where your data rules live: the minimum necessary information goes in, and nothing sensitive goes in without approval. Write that boundary into the brief and it gets applied every time, not just on careful days.
Part 04
Examples
What does good look like?
Definition
One concrete instance of the output you want: a past document that worked, a shape to follow, a filled-in sample.
Example
# EXAMPLE Follow the shape of the March update summary (attached): three sections — What changed · What it means · What to do.
Why it matters
One example replaces twenty adjectives. "Professional but friendly" is a guess about your taste; a sample is a specification of it.
How the output will be checked, written into the brief itself: what the assistant must surface, and what you will verify before the work ships.
Example
# EVALUATION List your assumptions and anything you could not confirm from the attachment. I will verify the effective dates and dollar amounts against the policy text before this goes out.
Why it matters
Asking for assumptions makes the remaining guesses announce themselves. Stating your own check plan keeps the last step human — a confident tone is not verification, and the tool's self-assessment is not either.
Where the metaphor stops
Brief it like a colleague — up to a point.
A colleague learns you. The assistant starts from zero each session, unless your tool stores context — know whether yours does, and what it keeps.
A colleague pushes back. An impossible ask gets questioned. Generated text attempts anything, fluently, including tasks that should have been refused.
A colleague notices gaps. "You forgot to mention the deadline" is a favour the assistant cannot do you. The brief must carry everything, every time.
This is why the structure matters more here than anywhere else at work. With a person, an incomplete brief gets repaired by the relationship. Here, the brief is the entire relationship.
Step 03 · The rebuild
Same ask, rebuilt as a brief.
Add the five parts in order and watch what each one removes. This is the request from Step 01 — same document, same tool.
Add the five parts in order — the note here explains what each one changes.
The request, rebuilt0 / 5 parts
Can you summarize the attached document for the team?
# Context
You are helping a team lead at a B.C. benefits carrier. The attached update changes how physiotherapy claims are assessed from 1 September. The team applies these rules daily.
# Objective
Produce a one-page summary the team can act on: what changed, what stays the same, what to do differently from 1 September. Done means a claims processor could apply the new rules after one read.
# Constraints
One page. Plain language, no legal phrasing. No member details or personal information. Where the update is ambiguous, flag it — do not guess.
# Example
Follow the shape of the March update summary (attached): three sections — What changed · What it means · What to do.
# Evaluation
List your assumptions and anything you could not confirm from the attachment. I will verify the effective dates and dollar amounts against the policy text before this goes out.
Step 04 · Try it
Check your read.
Three quick calls, each with feedback. Choose again until the reasoning clicks — nothing here is scored or stored.
Exercise 01Unsolved
Spot the missing part.
A colleague shows you this brief. It is close — one of the five parts is absent. Which one?
You are supporting a manager at a B.C. benefits carrier. Draft an email telling the team about the new intake process, under 150 words, plain language, no member information. Use the attached May announcement as the shape to follow.
Exercise 02Unsolved
Pick the constraint the work can obey.
Only one of these three is an instruction the output can actually be checked against.
Exercise 03Unsolved
Redirect the draft.
The first draft came back too formal, and the 1 September date is buried in the third paragraph. What is the strong next message?
Step 05 · The conversation
One brief, then a dialogue.
The first output is a fast first draft, not a verdict. Expect a short conversation — and know the habits that quietly ruin it.
Draft→Inspect→Redirect→Verify — human→Edit
Anti-pattern 01The vague askTen words of instruction, six silent guesses. You met it in Step 01.
Anti-pattern 02Prompt superstitionCopying "magic" wording without knowing which part does the work. Structure transfers; incantations do not.
Anti-pattern 03Data dumpingPasting everything "for context." Use the minimum necessary information — attention and privacy both have budgets.
Anti-pattern 04First-draft acceptanceShipping the initial output because it reads confidently. Fluency is not evidence.
Anti-pattern 05Outsourced judgmentLetting the tool decide what matters, what is fair, or what is true. The brief delegates work — never accountability.
And sometimes the strong move is no brief at all. Skip the assistant when the data is not approved for the tool, when the stakes outweigh the time saved, or when a deterministic tool — calculator, template, search, the policy document itself — gives an exact answer. Choosing the right instrument is part of the fluency.
Tool notes · Captured 27 July 2026
In Microsoft 365 Copilot, this maps cleanly.
Microsoft's own structure
Copilot guidance uses GCSE — goal, context, source, expectations — and only the goal is required. Same bones as the five parts: goal is the objective; expectations carry constraints and examples; source is context you attach rather than describe.
Ground it in sources
Reference the actual files, meetings, and mail with the / picker instead of describing them from memory. Most unexpected output traces to a missing source boundary — the tool answered from general knowledge when you meant a specific document.
Match the mode to the stakes
Reasoning modes trade speed for deliberation. Quick modes suit routine asks; deeper-reasoning modes suit multi-step, regulated, or high-consequence work. Choose by consequence, not by habit.
Menus move
Model pickers, mode names, and defaults change over product cycles — details here carry this card's capture date. The five-part brief is the part that transfers.
Concepts transfer; interfaces move. Learn the brief once and you can sit down at Copilot, Claude, or Gemini and be good at all of them by lunch.
Step 06 · Send
Send the brief. Then check the work.
The brief gets you a strong draft — not a finished one.
The five parts put your judgment into the request; the review puts it into the result. That second half has its own job aid: run The Human Check before anything you built here leaves your hands. Print this page and the paper version is your workshop worksheet — the brief canvas, the quick reference, and the watchlist, ready for a pen.
Lineage
Developed from "CoPilot AI Augmentation & Prompting Guide," a quick-reference job aid written for a colleague (June 2026), generalized for public use.
Educational guidance, not policy. Where your organization sets rules for AI use, approved tools, or data handling, those rules govern every brief you write — this page never overrides them.