| # | Pattern | Template |
|---|---|---|
| 1 | Give a role | “You are an experienced [role] helping [audience].” |
| 2 | State the goal and audience | “Write this for [reader] who needs to [decision/action].” |
| 3 | Provide context | “Here is the background: …” |
| 4 | Separate input clearly | “Use only the text between the triple quotes.” |
| 5 | Show examples | “Here are two examples of the output I want: …” |
| 6 | Specify the format | “Answer as a table with columns A, B, C.” |
| 7 | Set constraints | “Max 120 words. Plain language. No jargon.” |
| 8 | Ask for steps | “Reason step by step, then give the final answer.” |
| 9 | Allow uncertainty | “If you are not sure, say so instead of guessing.” |
| 10 | Ask for sources/quotes | “Support each claim with a quote from the document.” |
| 11 | Critique then rewrite | “List the 3 main weaknesses, then rewrite to fix them.” |
| 12 | Ask what’s missing | “What information would you need to do this better?” |
Combine them
A strong prompt usually uses 3–5 patterns at once:
You are an editor for a technical blog (1). Rewrite the draft between the quotes for developers new to the topic (2), as three short sections with headings (6), under 300 words (7). If something in the draft seems wrong, flag it instead of fixing it silently (9).
Iterate
Treat the first answer as a draft. The fastest improvements come from telling the model what was wrong (“too formal”, “missing the pricing”) rather than starting over.
More templates in our Prompt Library.
Before and after examples
Vague: “Write about our new product.”
Better (role + context + format): “You are a product marketer. Write a 120-word announcement of our new project-management app for small agencies. Highlight the shared calendar and client portal. Friendly, no hype. End with a call to action to start a free trial.”
Vague: “Explain inflation.”
Better (audience + examples + check): “Explain inflation to a 14-year-old using an example about the price of snacks. Then ask me two questions to check I understood.”
Vague: “Fix my code.”
Better (context + constraints): “This Python function should return the average of a list but fails on empty lists. Fix it so it returns None for empty input, keep the same function signature, and explain the change in one sentence.”
Patterns to avoid
- Stacking contradictory instructions (“be very detailed” and “keep it short”).
- Hiding the real question at the start of a long paste — put the task last, or repeat it.
- Asking for facts without sources when accuracy matters.
- Accepting the first answer — iteration usually improves results quickly.
A reusable prompt skeleton
Role: You are a [expert type]. Context: [background the model needs]. Task: [what to do, in one sentence]. Constraints: [length, tone, what to avoid, what to keep]. Format: [bullets, table, JSON, email…]. Examples: [optional input/output pairs]. Check: Before answering, verify [specific criterion]; if information is missing, ask me.
Fill in only the parts you need. Save versions that work for your recurring tasks.
Pattern quick reference
| Pattern | Use it when | Example phrase |
|---|---|---|
| Role | You need expert framing | “You are an experienced HR manager…” |
| Context | The model lacks background | “Our company sells… to…” |
| Examples | You want a specific style or format | “Here are two examples of the tone I want…” |
| Step by step | The task needs reasoning | “Work through this step by step, then give the answer.” |
| Format | You need structured output | “Answer as a table with columns…” |
| Constraints | You need limits | “Under 100 words, no jargon.” |
| Critique | You want quality control | “Review your answer against these criteria and fix issues.” |
| Ask first | Requirements are unclear | “Ask me up to three questions before starting.” |
Most good prompts combine three or four of these patterns. Start simple, check the result, and add patterns only where the output falls short — piling on every pattern at once makes prompts harder to debug.
What to watch next
- Assistants that ask clarifying questions automatically.
- Saved prompt templates and custom instructions in AI tools.
- Reasoning models that need less step-by-step prompting.
- Prompt libraries shared within organizations.
Key takeaways
- Specify role, context, task, format and constraints for better results.
- Give examples when you want a specific style or structure.
- Ask for step-by-step reasoning or verification on complex tasks.
- Combine patterns and iterate rather than expecting perfection first time.
Frequently asked questions
What is the most useful prompting pattern?
Giving clear context and the desired format usually improves results the most for everyday tasks.
Do prompting patterns work with every AI model?
The core patterns work with all capable chat models, though the best wording can vary between models.
Should I use very long prompts?
Include the information the task needs, but remove irrelevant material. Clarity matters more than length.