Course resource
PCTC Prompt Canvas
The four-part structure behind every prompt that reliably works. Fill the canvas, then assemble.
The four parts
P — Persona. Who should the AI be? This sets vocabulary, assumptions, and what it treats as obvious.
C — Context. What does it need to know that it cannot infer? Audience, constraints, background, what came before.
T — Task. What exactly do you want produced? One verb, one deliverable.
C — Constraints. Length, format, tone, and what not to do.
Most bad prompts are missing Context and Constraints. Most people write only the Task and wonder why the output is generic — a generic prompt has no way to produce a specific answer.
The canvas
Fill this in before writing anything.
| Part | Question to answer | Yours |
|---|---|---|
| Persona | Whose expertise do I want? | |
| Context | Who is the audience? | |
| What background is needed? | ||
| What have we already decided? | ||
| Task | What one thing should it produce? | |
| What does "done" look like? | ||
| Constraints | How long? | |
| What format? | ||
| What tone? | ||
| What must it avoid? |
Assembling it
You are [PERSONA].
CONTEXT
[AUDIENCE]
[BACKGROUND]
[WHAT IS ALREADY DECIDED]
TASK
[ONE SENTENCE, ONE VERB]
CONSTRAINTS
- Length: [LENGTH]
- Format: [FORMAT]
- Tone: [TONE]
- Do not: [WHAT TO AVOID]
Worked example
Weak:
Write a follow-up email to a client.
Canvas filled:
- Persona: an account manager who has run this relationship for two years
- Context: client is a mid-size manufacturer; they went quiet after we sent a quote three weeks ago; the quote was at the top of their stated budget; their procurement lead is new
- Task: draft a follow-up that reopens the conversation
- Constraints: under 120 words; no attachments; warm but not needy; do not discount; do not ask "just checking in"
Assembled:
You are an account manager who has run this client relationship for two years.
CONTEXT
The client is a mid-size manufacturer. They went quiet three weeks ago after
we sent a quote at the top of their stated budget. Their procurement lead is
new and was not part of the earlier conversations.
TASK
Draft a follow-up email that reopens the conversation.
CONSTRAINTS
- Under 120 words
- Warm but not needy
- Do not offer a discount
- Do not use "just checking in" or "circling back"
- Give the new procurement lead a reason to reply
The second one produces something you might actually send. The difference is not length — it is that every sentence of the prompt removes a wrong answer.
Negative space
Telling AI what not to do is often more effective than telling it what to do, because it removes the defaults you did not ask for.
Useful constraints to keep in your back pocket:
- Do not open with a summary of my question.
- Do not use "delve", "leverage", "robust", "seamless", "landscape".
- Do not hedge. If you are uncertain, say so once and continue.
- Do not add a concluding paragraph that restates the above.
- Do not use bullet points unless the content is genuinely a list.
- Do not invent statistics, dates, or citations. Write [VERIFY] instead.
When the output is wrong
Diagnose before rewriting the whole thing.
| What is wrong | Which part to fix |
|---|---|
| Right content, wrong vocabulary | Persona |
| Generic, could apply to anyone | Context |
| Answered a different question | Task |
| Right idea, wrong shape or length | Constraints |
| Confidently wrong facts | None of them — add a verification step |
Change one part at a time. Changing all four tells you nothing about which one was the problem.
Your saved canvases
Once a prompt works, save it with the canvas filled in, not just the final text. When you need to adapt it six months later, the canvas tells you which part to change.
| Prompt name | Persona | Used for | Where saved |
|---|---|---|---|