Reference Guide

14 Prompt Engineering Frameworks

Understand which framework fits your task — pros, cons, and when to use each methodology.

14 of 14

CRISPE

Capacity, Role, Insight, Statement, Personality, Experiment

Try it →

Strengths

Most comprehensive framework — covers every context dimension
Excellent for persona-driven and creative tasks
Personality field produces distinctive, consistent AI voice

Limitations

Six fields is heavy for simple tasks
Easy to over-engineer straightforward requests

Best forComplex content generation, persona-driven writing, creative briefs

Avoid whenQuick factual queries or single-step tasks

CO-STAR

Context, Objective, Style, Tone, Audience, Response

Try it →

Strengths

Strong tone and audience control — ideal for communication work
Style + Tone combination produces polished, consistent copy
Clear objective field keeps AI focused on deliverables

Limitations

Weak for technical or analytical tasks
Context and Objective fields often overlap

Best forMarketing copy, blog posts, communication drafts, audience-targeted content

Avoid whenCode generation, data analysis, structured reasoning

RISEN

Role, Instructions, Steps, End Goal, Narrowing

Try it →

Strengths

Step-by-step structure prevents AI drift on long tasks
Narrowing field eliminates irrelevant output
End Goal keeps every step purposeful

Limitations

Requires knowing the process steps upfront
Less useful for exploratory or open-ended tasks

Best forSOPs, tutorials, process documentation, instructional guides

Avoid whenOpen-ended research, brainstorming, discovery tasks

RTCF

Role, Task, Context, Format

Try it →

Strengths

Four focused fields — fast to fill, hard to leave out key info
Context field prevents generic outputs without requiring a full framework
Great default starting point for any structured task

Limitations

No tone, audience, or step-by-step control
Less suited for multi-part reasoning or creative nuance

Best forEveryday business tasks, drafts, and structured deliverables

Avoid whenComplex reasoning chains or tasks requiring explicit tone/persona control

Chain of Thought

Task, Reasoning Style, Output Format

Try it →

Strengths

Dramatically improves accuracy on reasoning-heavy tasks
Forces AI to show its work — errors are visible and fixable
Proven to reduce mistakes in multi-step logic

Limitations

Produces verbose outputs — not suitable for brief responses
Slower and costlier in production AI pipelines

Best forMath problems, logical deduction, debugging, multi-step analysis

Avoid whenSimple factual questions, creative writing, when brevity matters

Few-Shot

Task, Examples, Output Format

Try it →

Strengths

Most reliable output format control — AI learns by example
Teaches formatting, tone, and pattern simultaneously
Works even when instructions alone would be ambiguous

Limitations

Requires crafting quality examples before you can use it
Long prompts increase token cost

Best forClassification, data extraction, consistent formatting, pattern replication

Avoid whenNovel tasks with no clear examples, purely open-ended generation

APE

Action, Purpose, Expectation

Try it →

Strengths

Ultra-concise — forces clarity of intent
Fast to fill, fast to iterate
Expectation field provides built-in success criteria

Limitations

Minimal context leads to generic outputs on complex tasks
No role, tone, or step control

Best forSimple well-defined tasks, quick one-off requests

Avoid whenComplex multi-part tasks, when background context matters

ReAct

Task, Reasoning Steps, Action Format, Observation Format

Try it →

Strengths

Best framework for agentic AI — built for tool use and multi-step reasoning
Observation loop enables error correction mid-task
Maps directly to LangChain, AutoGPT, and similar agentic systems

Limitations

Complex to set up — requires understanding agentic AI concepts
Overkill for standard content or analysis tasks

Best forAI agents, tool-using systems, autonomous task completion, code execution loops

Avoid whenSimple Q&A, content generation, anything without tool use

BROKE

Background, Role, Objectives, Key Results, Evolve

Try it →

Strengths

OKR-style structure brings business clarity to AI tasks
Key Results field forces measurable outcomes
Evolve field enables iterative refinement within one prompt

Limitations

Business jargon can confuse AI on non-business tasks
Overkill for anything outside strategy or planning

Best forBusiness strategy, project planning, goal-setting, OKR workshops

Avoid whenTechnical tasks, creative work, casual or personal queries

RODES

Role, Objective, Details, Examples, Sense Check

Try it →

Strengths

Built-in quality control via Sense Check — rare among frameworks
Examples field strengthens output consistency
Details field allows nuanced constraint specification

Limitations

Sense Check is a novel concept that AI may not always leverage effectively
Five fields is heavyweight for simple tasks

Best forHigh-stakes outputs, detailed analysis, tasks where accuracy is critical

Avoid whenQuick low-stakes content, simple formatting tasks

TRACE

Task, Role, Audience, Create, Expectation

Try it →

Strengths

Audience-first design produces highly targeted communication
Create field separates deliverable type from content instructions
Good for educational and instructional content

Limitations

Overlaps with CO-STAR in many areas
Less differentiated — CO-STAR or CRISPE often outperform it

Best forAudience-specific content, educational material, targeted messaging

Avoid whenInternal technical documentation, tasks without a clear end audience

CARE

Context, Action, Result, Example

Try it →

Strengths

Result-driven framing keeps AI focused on outcomes not just actions
Example field reinforces expected output format
Concise — four fields covers most task dimensions

Limitations

Limited persona definition compared to CRISPE or RISEN
Context field alone may not be enough for complex backgrounds

Best forOutcome-focused business writing, task-specific requests, professional emails

Avoid whenComplex multi-step processes, deep persona tasks

Structured Output

Task, Output Format, Schema, Constraints

Try it →

Strengths

Best framework for forcing specific data formats and schemas
Essential for API integrations and data pipelines
Constraints field prevents hallucinated fields

Limitations

Purely functional — no style, persona, or tone control
Rigid structure makes it unsuitable for narrative content

Best forJSON extraction, data pipelines, structured data generation, API response formatting

Avoid whenCreative writing, conversational tasks, any output needing prose

Auto-CoT

Task, Number of Sub-questions, Final Answer Format

Try it →

Strengths

Self-directed reasoning — AI generates its own sub-questions
Reduces prompt engineering effort for exploratory tasks
Good for tasks where you do not know the reasoning path

Limitations

Less control over reasoning direction — AI may go off-course
May generate irrelevant sub-questions without careful task framing

Best forResearch tasks, exploratory analysis, open-ended problem solving

Avoid whenPrecise step-by-step processes where you need to control the reasoning flow