You'll leave with: a reusable template for clear prompts and the ability to fix a weak prompt in under two minutes.
The big idea: brief a smart new intern
Anthropic's prompting guide offers a handy test: show your prompt to a colleague with minimal context on the task and ask them to follow it. If they'd be confused, the AI will be too. The model knows a lot about the world but nothing about your situation except what's on the page.
So don't write a wish. Write a brief. We use six parts, easy to remember as R-C-T-F-E-C:
| Part | Question it answers | Example |
|---|---|---|
| Role | Who should it act as, and for whom? | “You are a friendly personal-finance explainer for beginners.” |
| Context | What does it need to know about my situation? | “I'm planning dinner for 4. Budget $40. No peanuts.” |
| Task | What exactly should it do, with a verb? | “Write a grocery list and total it.” |
| Format | What should the answer look like? | “A table: item | qty | price. Then a one-line total.” |
| Examples | What does a good answer look like? | One or two sample rows or a sample sentence |
| Constraints | What must it avoid, and what are the limits? | “Under 150 words. If unsure, say so. Don't invent prices.” |
Before and after
Weak prompt
Help me with my budget.
The AI has to guess everything: your income, your goal, the format, how long you want it.
Strong prompt
Role: You are a friendly personal-finance coach for beginners. Context: I earn about [AMOUNT] a month after tax. My fixed costs are [LIST]. I want to save [GOAL] by [DATE]. Task: Build a simple monthly budget and tell me where I'm most likely overspending. Format: A table with category | amount | % of income, then 3 bullet-point suggestions. Example of a row: "Groceries | $400 | 18%". Constraints: Use my numbers only. If you need a number I didn't give you, ask. Under 200 words. This is general information, not financial advice.
Each part, in more detail
Role
One sentence is enough. Anthropic's guide notes a role helps focus tone and behaviour, and its 2026 prompt article warns not to over-constrain it: “helpful assistant” often beats “world-renowned expert who never makes mistakes”. A better role-equivalent is often a perspective: “explain it for someone who has never used a spreadsheet”.
Context
Give facts, not adjectives. “Small budget” is vague; “$40 for four people” is useful. Put long pasted material in a clearly marked block so the AI can tell your data from your instructions:
Using only the text between the <<< and >>> markers, list the 5 key points. <<< [PASTE TEXT HERE] >>>
Anthropic's guide suggests XML-style tags for complex prompts; its newer article says clear headings and plain language work as well for most cases. Pick a style and stay consistent.
Task
Lead with an action verb: write, compare, list, rewrite, check. One main task per prompt. If you need three outputs, number them.
Format
Say it: table, bullet list, 3 sentences, email with subject line, JSON. Ask for a length. This also helps with tokens (Lesson 5): shorter answers cost less time and usage.
Examples
Examples are one of the most reliable ways to steer format, tone and structure, according to Anthropic's guide, which suggests relevant, varied examples (it recommends 3-5 for best results on complex formats). For everyday prompts, one good example is often enough. Show the shape you want, and say “match this style, not these exact words”.
Constraints
This is where you stop the AI from being confidently wrong: “Don't invent sources.” “If you aren't sure, say 'I'm not sure'.” “Don't use jargon.” “Ask me up to 3 questions before you start if something is missing.” That last one is the cheapest quality boost there is.
The reusable template
Role: You are [ROLE] helping [WHO]. Context: [FACTS ABOUT MY SITUATION: numbers, constraints, what I already tried]. Task: [ACTION VERB] [WHAT]. Format: [TABLE / BULLETS / N SENTENCES / EMAIL], under [N] words. Example (match the style, not the words): [ONE SAMPLE]. Constraints: [WHAT TO AVOID]. If you're unsure, say so. If anything is missing, ask me up to 3 questions first.
Exercise
Exercise A: Fix a weak prompt. Rewrite each of these using the template. Aim for 4 lines or fewer each.
- “Write something about my business.”
- “Explain taxes.”
- “Make a workout plan.”
Exercise B: Test one part at a time. Take a prompt you use often. Run it three times: (1) as-is, (2) with a Format added, (3) with a Constraint added. Which change helped most? Write it on a sticky note: that's the part you usually skip.
Exercise C: Intern test. Read your best prompt out loud to a friend. Do they ask “what do you mean by…?” Every question they ask is a missing piece of context.
Key takeaways
- Brief the AI like a smart new intern: it only knows what's on the page.
- Role, Context, Task, Format, Examples, Constraints. Skip parts you don't need, but never skip Task and Constraints on important prompts.
- “Ask me up to 3 questions first” is the cheapest upgrade you can make.
Related: Browse the Prompt Library · Try the Explain-It-Like-a-Human Tutor
Sources
Checked November 10, 2026 (Eastern Time). Plans, limits and features change: confirm on the official page.