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

Write me a marketing plan for our new product launch.

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

Role: You are a senior product marketing manager with 12 years of experience launching B2B SaaS products, specializing in developer tools and engineering team workflows.

Instructions: Create a go-to-market plan for our AI-powered code review tool targeting engineering teams at mid-market companies (50 to 200 employees).

Steps:
1. Identify our top three differentiators versus competitors like CodeClimate, SonarQube, and Codacy.
2. Define two buyer personas (engineering manager and VP of engineering) with their pain points and decision criteria.
3. Outline a 60-day launch timeline with three phases: pre-launch, launch week, and post-launch momentum.
4. Recommend four marketing channels ranked by expected ROI with budget allocation.
5. Draft three key messaging statements connecting product features to business outcomes.

End Goal: Produce a plan the marketing team can execute immediately, with clear ownership for each phase and measurable KPIs for every milestone.

Narrowing: Focus on North American market only. Budget is $30K for Q1. Write for a non-technical audience (engineering leadership will review). Keep under 2,000 words with headers, bullet points, and a summary table.

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?"

Role: You are a senior customer insights analyst with 8 years of experience in SaaS, specializing in qualitative feedback analysis and NPS program design.

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.

Instructions: Analyze the following 50 customer feedback comments and produce a prioritized list of product improvement opportunities.

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.

Steps:
1. Categorize each comment by theme (usability, performance, missing features, pricing, support).
2. Count frequency and identify the top five themes by volume.
3. For each top theme, pull two representative quotes.
4. Assess business impact of each theme using a high/medium/low scale.
5. Rank improvement opportunities by combining frequency and business impact.

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.

End Goal: The product team should be able to use this analysis in their next sprint planning session to prioritize the top three improvements with the greatest impact on customer retention.

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

Narrowing: Present the analysis as a structured report with a summary table followed by detailed sections for each theme. Keep it under 800 words. Use a data-driven tone without subjective opinions. Do not suggest solutions; only identify and rank the problems.

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

Role: You are a performance marketing specialist with 8 years of experience running paid acquisition campaigns for B2B SaaS companies, particularly skilled at LinkedIn and Google Ads.

Instructions: Design a paid advertising campaign to generate 200 qualified demo requests for our HR analytics platform in 60 days.

Steps:
1. Define the ideal customer profile and three audience segments for targeting.
2. Create ad copy variations for LinkedIn (sponsored content and message ads) and Google Search.
3. Design a landing page structure with headline, subheadline, three benefit blocks, and CTA.
4. Set up a budget allocation plan across channels and audience segments.
5. Define the measurement framework with conversion tracking at each funnel stage.

End Goal: Deliver a campaign plan that the paid media team can launch within one week, with all creative briefs, targeting parameters, and budget splits ready to execute.

Narrowing: Total budget is $25K. Target companies with 200 to 2,000 employees in the US and Canada. Write ad copy at an 8th-grade reading level. Keep the full plan under 2,000 words. No video ad recommendations for this phase.

Example 2: Code Review Analysis

Role: You are a principal software engineer with 15 years of experience in Python and distributed systems, serving as the technical lead for a team of 12 engineers.

Instructions: Review this Python codebase for a FastAPI microservice that handles payment processing and provide actionable improvement recommendations.

Steps:
1. Evaluate the overall architecture and identify any violations of SOLID principles.
2. Check error handling patterns, particularly around external API calls and database transactions.
3. Assess the test coverage and identify critical paths that lack tests.
4. Review security practices including input validation, secret management, and SQL injection prevention.
5. Flag any performance bottlenecks, especially in database queries and async operations.

End Goal: Produce a review document that junior and mid-level engineers can use to improve the codebase over the next two sprint cycles, with each issue prioritized by severity and effort.

Narrowing: Focus only on the payment processing module, not the full application. Categorize issues as Critical, High, Medium, or Low. Provide specific code examples for each recommendation. Keep the review under 2,500 words. Do not suggest rewriting the entire service.

Example 3: Data Analysis Report

Role: You are a senior data analyst with expertise in e-commerce analytics and customer behavior modeling, with a background in statistical analysis using Python and SQL.

Instructions: Analyze our Q1 2026 e-commerce performance data and identify the three most impactful opportunities for revenue growth in Q2.

Steps:
1. Break down revenue by channel (organic, paid, email, referral) and compare to Q4 2025.
2. Analyze customer cohort data to identify which segments have the highest lifetime value.
3. Examine the conversion funnel from landing page to purchase and pinpoint the largest drop-off point.
4. Cross-reference product category performance with margin data to find high-margin growth opportunities.
5. Synthesize findings into three prioritized recommendations with estimated revenue impact.

End Goal: The analysis should give the VP of Marketing a clear, data-backed case for where to allocate the Q2 budget, with specific dollar amounts tied to each recommendation.

Narrowing: Use only the data points provided (do not fabricate numbers). Present findings with tables and bullet points. Write for a business audience, not a technical one. Keep the analysis under 1,800 words. Include a one-paragraph executive summary at the top.

Example 4: Education and Training

Role: You are an instructional designer with 10 years of experience creating corporate training programs, specializing in technical onboarding for software companies.

Instructions: Design a 4-week onboarding curriculum for new junior developers joining our backend engineering team.

Steps:
1. Map the core competencies a junior developer needs to be productive on our stack (Python, Django, PostgreSQL, Docker, AWS).
2. Structure the four weeks with daily learning objectives and hands-on exercises.
3. Create assessment checkpoints at the end of each week to measure progress.
4. Design a mentorship pairing system with specific discussion prompts for mentor-mentee sessions.
5. Build a resource library with documentation, video tutorials, and practice projects for self-paced learning.

End Goal: New hires should be able to complete their first code review and submit their first pull request by the end of week 3, with week 4 focused on their first feature assignment.

Narrowing: Assume new hires have CS degrees but no professional experience. Limit daily learning time to 6 hours (2 hours mentor-led, 4 hours self-paced). Include a simple tracking spreadsheet template. Keep the curriculum document under 3,000 words. Do not assume access to paid learning platforms.

Example 5: Business Plan Section

Role: You are a startup advisor with 20 years of experience mentoring seed-stage founders, specializing in marketplace and platform business models.

Instructions: Create a one-page business plan summary for a freelance developer marketplace that connects companies with vetted, specialized developers for short-term projects.

Steps:
1. Define the value proposition for both sides of the marketplace (companies and developers).
2. Outline the revenue model with specific pricing tiers and take rates.
3. Identify the top three risks and mitigation strategies for each.
4. Map the first 12 months with quarterly milestones and key metrics.
5. Summarize the funding ask with a use-of-funds breakdown.

End Goal: This summary should be compelling enough for an angel investor to request a full pitch deck and a 30-minute meeting.

Narrowing: Strictly one page (approximately 500 to 600 words). Use bullet points and headers only. Do not include financial projections beyond Year 1. Focus on the North American market. Write in a confident but not hyperbolic tone.

Example 6: Technical Documentation

Role: You are a senior technical writer with 8 years of experience writing API documentation for developer platforms, with expertise in REST and GraphQL APIs.

Instructions: Write the getting-started guide for our webhook notification API that developers will use to receive real-time event updates from our platform.

Steps:
1. Write a two-paragraph overview explaining what webhooks are and why they are useful.
2. Document the setup process: registering an endpoint, selecting event types, and configuring retry policies.
3. Provide a complete code example in Python showing how to receive and validate a webhook payload.
4. Explain the security model including signature verification and IP whitelisting.
5. Add a troubleshooting section covering the five most common integration issues.

End Goal: A developer with intermediate experience should be able to set up their first webhook integration and receive test events within 30 minutes of reading this guide.

Narrowing: Write for developers who know HTTP but may not have used webhooks before. Include code examples in Python only (other languages will be separate guides). Keep the guide under 2,000 words. Use standard documentation format with numbered steps and code blocks. Avoid internal jargon.

Example 7: Creative Writing Brief

Role: You are an experienced brand copywriter with a portfolio of work for direct-to-consumer lifestyle brands, specializing in email marketing that drives conversions.

Instructions: Write a 5-email welcome sequence for new subscribers to our premium coffee subscription service.

Steps:
1. Email 1 (Day 0): Welcome and brand story to establish trust and personality.
2. Email 2 (Day 2): Product education covering the sourcing process, roast profiles, and what makes our coffee different.
3. Email 3 (Day 5): Social proof with customer testimonials, press mentions, and subscriber count.
4. Email 4 (Day 8): Objection handling addressing common hesitations like cost, commitment, and taste preferences.
5. Email 5 (Day 12): Conversion push with a limited-time offer, clear CTA, and urgency.

End Goal: The sequence should convert at least 15% of new subscribers into paying customers within the first 14 days, building a relationship that reduces churn beyond the first shipment.

Narrowing: Each email should be 150 to 250 words. Write in a warm, conversational tone that feels like a friend who loves coffee, not a salesperson. Subject lines must be under 50 characters. Include one CTA per email. Do not use discount percentages above 20%. Format each email with subject line, preview text, body, and CTA button text.

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

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.