RISEN Framework: 5 Steps to Better AI Prompts

The RISEN Framework: A Complete Guide to Structured AI Prompts

Most AI prompts fail for the same reason: they leave too much for the model to guess. You type a sentence or two, hit enter, and get back something vague, overlong, or completely off-target. RISEN framework prompt engineering solves this by giving you a five-part structure that tells the AI exactly who to be, what to do, how to do it, what success looks like, and what boundaries to respect. Created by Kyle Balmer as an evolution of the RISE framework, RISEN has become one of the most practical prompt engineering methods for anyone working with ChatGPT, Claude, or other large language models.

I have used RISEN across hundreds of prompts for business strategy, technical documentation, content creation, and data analysis. The difference is consistent: RISEN prompts produce outputs you can actually use on the first try, while unstructured prompts produce outputs you spend 20 minutes reworking. This guide walks you through each component, shows you seven complete examples you can copy and adapt today, and compares RISEN to other popular frameworks so you can choose the right tool for each task.

For the quick-reference version of the framework itself, see the RISEN framework page. This tutorial goes deeper with step-by-step construction, common mistakes, and model-specific tips.

What Is the RISEN Framework?

The RISEN framework is a five-component prompt structure created by Kyle Balmer as an evolution of the RISE framework. Where RISE gives you Role, Instruction, Specifics, and Examples, RISEN replaces the last two components with Steps, End Goal, and Narrowing, shifting the focus from describing desired output to directing the AI through a specific process.

Here is what each component does:

The five components work together like a project brief. Role tells the AI who it should be. Instructions tell it what to produce. Steps tell it how to work through the task. End Goal tells it why the work matters. Narrowing tells it what boundaries to respect.

The key insight behind RISEN is that each component addresses a different failure mode. Role prevents generic, surface-level responses. Instructions prevent ambiguity about the deliverable. Steps prevent the AI from skipping important phases. End Goal keeps everything aligned with a real outcome. Narrowing eliminates the most common complaints: too long, wrong tone, off-topic tangents.

Why RISEN Works Better Than Simple Prompts

The best way to understand RISEN's value is to see the difference side by side.

Before: Simple Prompt

A typical AI response to this prompt is a generic 500-word overview with surface-level advice about social media, email marketing, and "identifying your target audience." Nothing specific. Nothing actionable. You spend the next 20 minutes re-prompting to get closer to what you actually needed.

After: RISEN Prompt

The RISEN prompt produces a structured, actionable document with named personas, a phased timeline, ranked channels with dollar amounts, messaging pillars tied to developer pain points, and measurable KPIs. First draft, ready to use, no re-prompting needed.

Step-by-Step: Building a RISEN Prompt From Scratch

Here is the process I follow every time I write a RISEN prompt. I will walk through constructing one for a customer feedback analysis.

Step 1: Define the Role

Ask yourself: "If I were hiring a human expert for this task, what would their title and specialization be?"

The more specific the role, the better the output. "Data analyst" is too broad. "Senior customer insights analyst specializing in qualitative feedback analysis" activates a far more relevant knowledge domain.

Formula: [Seniority level] + [job title] + [years of experience] + [specialization]

Step 2: Write the Instructions

Condense your entire request into a single sentence. Use a precise verb: analyze, create, draft, audit, compare.

If you cannot fit the task into one sentence, your request might need to be split into multiple prompts.

Step 3: Map Out the Steps

Think about how you would explain this task to a new team member. What steps would you walk them through? Number each step in logical order.

Aim for 3 to 6 steps. Fewer than three means you are not adding much structure. More than six usually means you are overcomplicating the task.

Step 4: Define the End Goal

Shift from "what should the output contain" to "what should the output enable." The End Goal is about outcomes, not outputs.

Formula: "This [deliverable] should enable [who] to [do what] by [when/how]"

Step 5: Add Narrowing Constraints

Cover at least three of these dimensions:

• Length: Word count, page count, or section count
• Format: Bullet points, tables, headers, paragraphs
• Tone: Formal, conversational, technical, simple
• Scope: What to include and what to exclude
• Audience: Who will read this and what they already know

Combine all five components and you have a prompt that produces exactly what you need on the first try.

7 Real-World RISEN Framework Examples

Here are seven complete RISEN prompts across different professional domains. Each one is tested and ready to customize. For more prompt templates, check out our best ChatGPT prompts collection and advanced prompt engineering techniques.

Example 1: Marketing Campaign Brief

Example 2: Code Review Analysis

Example 3: Data Analysis Report

Example 4: Education and Training

Example 5: Business Plan Section

Example 6: Technical Documentation

Example 7: Creative Writing Brief

RISEN vs RACE vs COSTAR vs RISE: Which Framework Should You Use?

RISEN is not the only structured prompting framework worth knowing. Here is how it stacks up against three popular alternatives. For a broader comparison, check out our guide on the best AI prompt frameworks in 2026.

Quick Decision Guide

• Simple one-shot tasks: Use RISE for speed and simplicity
• Expert-level analysis or consultation: Use RACE when role depth matters most
• Content where tone and audience are critical: Use COSTAR for writing tasks with strict voice requirements
• Complex multi-step tasks with constraints: Use RISEN when you need process control and output boundaries
• Goal-oriented prompts for beginners: Use SMART if you already know goal-setting frameworks
• Analytical reasoning tasks: Use CHAIN for debugging, math, and step-by-step analysis
• Quick tasks needing guardrails: Use APE for straightforward prompts

The biggest difference between RISEN and RACE is the Steps component. RACE tells the AI what to deliver and lets it figure out how. RISEN tells the AI what to deliver and how to get there. If your task has a specific workflow or methodology you want followed, RISEN is the better choice. If you want to give the AI freedom to approach the problem in its own way (as you would with a hired consultant), RACE is a better fit.

5 Common RISEN Mistakes (And How to Fix Them)

Mistake 1: Vague Steps That Do Not Guide the Process

Writing "Research the topic" as a step gives the AI no direction. It does not know how deep to go, what sources to consider, or what format to produce for that research phase.

Fix: Each step should specify the action, the scope, and the expected output. "Identify the top five competitors by market share and list their pricing tiers, target audience, and primary feature differentiators" is a step the AI can execute precisely.

Mistake 2: Skipping the End Goal Entirely

Many people provide Role, Instructions, Steps, and Narrowing but leave out the End Goal. The AI completes all the steps but produces something that misses the strategic point.

Fix: Always state what the output should enable. "This analysis should give the VP of Sales enough data to approve or reject the new pricing model this week" aligns every step with a real decision.

Mistake 3: Narrowing That Is Too Broad

Writing "keep it professional" as your only constraint is barely better than writing no constraint at all. "Professional" means different things in different contexts.

Fix: Include at least three specific constraints covering different dimensions: format (tables vs. prose), length (word count), tone (formal, conversational, technical), scope (what to include and exclude), and audience (who will read this).

Mistake 4: Role-Instruction Mismatch

Assigning the role of "senior financial analyst" but then asking for a blog post about team management creates conflicting signals. The AI does not know which expertise to prioritize.

Fix: Make sure your Instructions describe a task that the assigned Role would naturally perform. A financial analyst analyzes financial data. A content strategist writes articles.

Mistake 5: Too Many Steps for Simple Tasks

Using eight numbered steps for a task that only needs two or three creates unnecessary overhead. The AI spends token budget on trivial substeps instead of providing depth where it matters.

Fix: Match the number of steps to the complexity of the task. Three to five steps is the sweet spot for most tasks. Reserve six or more steps for genuinely complex, multi-phase projects.

Tips for Using RISEN With Different AI Models

ChatGPT (GPT-4, GPT-4o)

GPT-4 follows numbered steps reliably and tends to produce structured output that matches your Narrowing constraints closely. It responds well to explicit word count limits and format specifications. When using GPT-4, you can be very specific in your Steps component, and the model will follow the sequence faithfully.

Tip: GPT-4 sometimes adds unnecessary preambles like "Certainly!" or "Great question!" Add "Skip any introductory pleasantries and start directly with the deliverable" to your Narrowing section.

Claude (Claude 3.5 Sonnet, Claude 3 Opus)

Claude excels at the Role component and tends to maintain the assigned persona more consistently throughout long responses. It also handles nuance in the End Goal well, often making strategic connections between steps and the final objective that other models miss.

Tip: Claude performs especially well when your Narrowing component includes tone and audience details. It is also more likely to ask clarifying questions if your prompt has ambiguity, so be thorough in your Instructions to avoid back-and-forth. For more on getting the most from Claude, see our GPT-5 and GPT-4 prompting guide which also covers Claude comparisons.

Gemini (Gemini 1.5 Pro, Gemini Ultra)

Gemini handles multi-modal inputs well, so if your RISEN prompt includes references to images, documents, or data files, Gemini can incorporate them directly. It tends to be verbose, so explicit word count limits in Narrowing are especially important.

Tip: Gemini sometimes reorders your steps based on what it considers a more logical sequence. If step order matters, add "Follow these steps in the exact order listed" to your Instructions.

Frequently Asked Questions

What does RISEN stand for in prompt engineering?

RISEN stands for Role, Instructions, Steps, End Goal, Narrowing. It is a five-component prompt engineering framework created by Kyle Balmer that helps you structure AI prompts for complex tasks. Each component addresses a specific failure mode: Role prevents generic responses, Instructions prevent ambiguity, Steps prevent skipped phases, End Goal ensures strategic alignment, and Narrowing eliminates common output problems like wrong length or tone.

How is RISEN different from the RISE framework?

RISEN evolved from the RISE framework (Role, Instruction, Specifics, Examples) by replacing Specifics and Examples with three more actionable components: Steps, End Goal, and Narrowing. The key difference is that RISE asks you to provide example outputs, while RISEN asks you to define the process and constraints. This makes RISEN better suited for complex tasks where you cannot easily provide a sample of the desired output.

When should I use RISEN instead of RACE or COSTAR?

Use RISEN when your task has multiple phases that need to follow a specific order, and when you need tight control over the output format, length, and scope. Use RACE when the professional role is the most important factor and you want the AI to determine its own approach. Use COSTAR when audience awareness and tone are the most critical dimensions, such as marketing copy or customer-facing content.

Can I use RISEN with any AI model?

Yes. RISEN is model-agnostic because it addresses a universal problem: prompt ambiguity. The framework works with ChatGPT, Claude, Gemini, Llama, Mistral, and any other large language model. The structure is about how you communicate your request, not about exploiting model-specific features. That said, different models have different strengths with specific components, which is why the tips section above covers model-specific adjustments.

Start Using RISEN Today

The fastest path to better AI outputs is to stop typing the first thing that comes to mind and start using a structure. RISEN gives you five components that map directly to how you would brief a skilled colleague: tell them who to be, what to do, how to do it, what success looks like, and what boundaries to respect.

Here is how to begin:

• Pick one complex task you regularly use AI for.
• Write a RISEN prompt using the seven examples above as a starting point.
• Compare the output to what you normally get with unstructured prompts.
• Save your best RISEN prompts as reusable templates.

For the quick-reference framework page with a copy-paste template, visit the RISEN framework reference. For a side-by-side comparison of all major frameworks, see 9 Best AI Prompt Frameworks in 2026.