Best Prompt Frameworks in 2025: Templates and Examples

Prompt frameworks are checklists for talking to a model. They help you include the context, constraints, and format you would otherwise forget.

If you are newer to prompting, start with our Prompt Engineering in 2025 guide or this primer on how to write effective AI prompts.

If you want a shortcut, the AI Prompt Generator can turn any framework below into a clean prompt.

Quick picks

Need a strict output format: use RTF.
Need the model to plan before it answers: use APE.
Need guardrails plus a scoring rubric: use ROLE.
Need a steadier answer for a fuzzy task: use SxS.

CRISPE

What it is: Context, Role, Input, Steps, Parameters, Example.

Best for: multi step work where you care about constraints and consistency.

Template

Context: {{background}}
Role: You are {{role}}
Input: {{task or data}}
Steps: {{numbered steps}}
Parameters: {{constraints, style, length, format}}
Example: {{optional example output}}

Tip: If you keep getting almost right outputs, add a real example. Even a small one helps.

RTF (Role, Task, Format)

RTF is the fastest way to stop the model from wandering. It is also great when you need the answer in a specific shape, like JSON, a table, or a checklist.

Template

Role: You are {{role}}
Task: {{what to do}}
Format: {{exact output format}}

APE (Ask, Plan, Execute)

APE is useful when the problem is not hard, but it has a lot of steps. You ask the model to restate the goal, sketch a plan, then do the work.

Template

Ask: {{goal in one sentence}}
Plan: List the steps you will follow.
Execute: Do the task using the plan.

ROLE (Role, Objectives, Limits, Evaluation)

ROLE is what I use when I want predictable work from a prompt, especially for teams. The evaluation section acts like a rubric, so the model knows what done means.

Template

Role: {{role}}
Objectives: {{what success looks like}}
Limits: {{what not to do}}
Evaluation: {{how to score the output}}

If you work across models, the same rubric usually holds. For model specific notes, see our guides on Claude prompting and Gemini prompting.

SxS (Side by side selection)

SxS is simple: generate a few candidates, compare them, and pick one. It is a good fit for naming, positioning, and any task where there is no single right answer.

Template

Generate 3 options that take different approaches.
Then compare them side by side and pick the best.
Explain the choice in 3 to 5 sentences.

Copy and paste examples

Marketing brief with CRISPE

Context: We sell a prompt builder for small teams.
Role: You are a product marketer.
Input: Write a one page marketing brief for a landing page.
Steps:
1) Define target user and problem
2) List 3 key messages
3) Propose a simple page outline
Parameters: Use a friendly, direct tone. Keep it under 400 words.
Example: Use short headings and tight bullets.

Technical RFC with ROLE

Role: You are a staff engineer.
Objectives: Draft an RFC for adding audit logs to our app.
Limits: Do not invent product requirements. Call out open questions clearly.
Evaluation: The RFC must include scope, non goals, data model, API changes, rollout plan, and risks.

Analysis with APE plus SxS

Ask: Compare two onboarding flows and recommend one.
Plan: List the data you will use and the criteria.
Execute:
1) Summarize each flow
2) Generate 3 recommendations
3) Compare them side by side
4) Pick one and give next steps

When to use which

CRISPE: you want a full checklist, plus an example
RTF: you mainly need the right output shape
APE: you need the plan first
ROLE: you want guardrails and a rubric
SxS: you want options and a clear pick

FAQ

Do these frameworks work in ChatGPT, Claude, Gemini, and Grok? Yes. The structure carries over. If you are comparing models, this guide helps: Claude vs ChatGPT vs Gemini.

What should I do if the model ignores my format? Try RTF and make the format section concrete. If you need a refresher, read prompt engineering best practices.