RISEN Framework Prompting: Role, Instructions, Steps, End-goal, Narrowing

The risen framework gives you five labeled slots for any chat prompt: Role, Instructions, Steps, End-goal, and Narrowing. You stop burying the task inside a vague paragraph. You label who speaks, what to do, how to sequence work, what done looks like, and what to exclude. This page is a deep dive on RISEN only. It is not the CO-STAR versus CRAFT hub article. You leave with field definitions, a blank template, six paste-ready examples, a review checklist, and a soft path to https://promptmake.net/frameworks/risen when you want labeled output without memorizing headers. RISEN fits GPT-5.6 Sol, Claude Fable 5, and Gemini 3.5 Flash chat work where you need repeatable structure across a team.

What the risen framework is and who should use it

RISEN is a prompt skeleton. Each letter names one decision most people skip. Role sets voice and expertise level. Instructions state the core task in plain verbs. Steps break work into an ordered list the model can follow. End-goal describes the finished output so the model knows when to stop. Narrowing lists boundaries: length, tone bans, sources, audience limits, and things to omit.

Teams adopt RISEN when one-off prompts drift. Marketing, support, and ops groups paste the same five headers into a shared doc. New hires fill slots instead of copying a magic paragraph from a senior coworker. Solo builders use RISEN when a task fails twice and RTF or a single sentence is not enough structure.

RISEN pairs well with reasoning-class models when Steps stay short. You do not need "think step by step" on GPT-5.6 Sol or Claude Opus 5 for hard analysis. You need clear End-goal and Narrowing so the model does not over-deliver a novel when you wanted a table.

PromptMake at https://promptmake.net/frameworks/risen generates RISEN-labeled drafts from a rough idea. You still edit before production. The generator does not run your task or store prompts in a vault unless your org adds that separately.

The five RISEN fields explained

Treat each field as a contract line. If you cannot fill Narrowing with at least two real limits, the prompt will sprawl. If Steps has one item, merge it into Instructions unless order truly matters.

Copy this blank block into your doc and fill it once per recurring job:

Role: [voice, audience, expertise level]

Instructions: [primary task in one or two sentences]

Steps: [numbered sequence]

End-goal: [finished artifact description]

Narrowing: [length, tone, exclusions, source rules]

Below, each field gets detail and failure modes teams see in review.

Role: voice without cargo cult

Role is not a trophy title. "World-class expert" does not change output. Role should name who reads the result or how the model should sound: "internal finance analyst writing for a CFO," "patient tier-1 support agent," "technical editor for developers who know Python."

Swap test: if you replace Role with a different job title and nothing in the expected output changes, rewrite Role around audience or behavior. Audience-first roles beat credential stacking.

Keep Role under three lines. Long persona essays belong in system prompts for Custom GPTs or Gems, not in every user message.

Instructions and Steps: task versus sequence

Instructions hold the main verb: summarize, compare, draft, extract, critique, transform. One primary task per prompt. If you need two unrelated tasks, split into two prompts or chain them with explicit handoff in End-goal.

Steps list order when order changes quality: research then outline then draft; read file then list risks then rank. Use three to seven steps. More than ten steps often get skipped mid-list on fast chat models.

Do not duplicate Instructions inside every Step. Steps are milestones, not micro-prompts unless a step is genuinely complex.

End-goal and Narrowing: done and not-done

End-goal answers: what file, table, email, or section set should exist when the model stops? Name format: "Markdown table with five rows," "JSON array of objects with keys id and label," "Email under 120 words with subject line."

Narrowing is the guardrail field. Max words, banned phrases, no invented pricing, cite only pasted context, no medical claims, US English, no emojis. Narrowing prevents the model from adding a conclusion essay when you asked for bullets.

Teams that skip Narrowing get fluent drafts that fail legal or brand review. Spend sixty seconds here before you regenerate three times.

RISEN template you can paste today

Use this shell for any new job. Replace bracketed lines.

Role: [specific voice or reader]

Instructions: [verb + object + success criterion]

Steps:

1. [first milestone]
2. [second milestone]
3. [third milestone]

End-goal: [artifact + format]

Narrowing: [length, tone, exclusions, honesty rules]

Context: [paste source material below the RISEN block]

Context is not the sixth RISEN letter, but production prompts almost always need a Context block after the five fields. Put instructions before long pasted text so the model sees the contract first.

Paste-ready RISEN example 1: weekly status email

Role: Engineering manager writing a weekly update for a VP who scans email on mobile.

Instructions: Summarize progress, blockers, and next-week priorities for the payments squad based only on the notes below.

Steps:

1. List shipped items with dates.
2. List blockers with owner names from notes only.
3. List three priorities for next week.

End-goal: Email under 200 words. Sections: Shipped | Blockers | Next week. Bullet points only.

Narrowing: No invented metrics. If a date is missing, write DATE UNKNOWN. No blame language. US English.

Context: [paste standup notes]

Paste-ready RISEN example 2: PR security review

Role: Security reviewer for a backend team. Direct tone. Flag only high-confidence issues.

Instructions: Review the pasted diff for authentication and input-validation risks in the checkout service.

Steps:

1. Scan entry points and auth middleware usage.
2. List input paths that reach SQL or shell.
3. Rank findings by severity.

End-goal: Markdown with sections Critical | High | Medium. Each item: file path, line hint, issue, suggested fix in one sentence.

Narrowing: No findings without a specific code reference. If unsure, tag UNVERIFIED. Max 12 findings.

Context: [paste diff]

Paste-ready RISEN example 3: competitive one-pager

Role: Strategy analyst writing for executives with two minutes to read.

Instructions: Compare three named competitors for a workflow hub decision. Use only facts from Context and label gaps.

Steps:

1. State buyer criteria from Context.
2. Score each competitor on criteria in a table.
3. Write a recommendation paragraph with one clear pick.

End-goal: One-page brief. TL;DR three bullets, comparison table five rows max, recommendation 80 words max.

Narrowing: No invented pricing. Mark unknown cells NEEDS RESEARCH. Neutral tone. No vendor trash talk.

Context: [paste RFP notes and vendor links summary]

Paste-ready RISEN example 4: customer reply draft

Role: Support agent for a SaaS billing product. Empathetic, policy-aware.

Instructions: Draft a reply to a customer who was charged twice and wants refund timing.

Steps:

1. Acknowledge the duplicate charge.
2. State refund policy from Context.
3. Offer next step and escalation path.

End-goal: Email with Subject line plus body under 150 words. Sign-off: Support Team.

Narrowing: Do not admit fault without confirmation in Context. No legal promises. Plain language.

Context: [paste ticket and policy excerpt]

Paste-ready RISEN example 5: lesson plan outline

Role: Middle-school science teacher writing for substitute teachers.

Instructions: Build a 45-minute lesson outline on photosynthesis using only objectives from Context.

Steps:

1. List learning objectives.
2. Add timing per segment.
3. Add one hands-on activity and one check-for-understanding question.

End-goal: Markdown outline with H2 per segment. Total time 45 minutes.

Narrowing: No copyrighted worksheet text. Materials list under ten items. Reading level grade 7.

Context: [paste district objectives]

Paste-ready RISEN example 6: data extraction to JSON

Role: Data assistant for an ops team. Precise, no commentary.

Instructions: Extract meeting action items from the transcript into structured JSON.

Steps:

1. Identify lines that assign an owner and deadline.
2. Normalize dates to ISO-8601 when explicit.
3. Output valid JSON only.

End-goal: JSON array of objects with keys task, owner, due_date. due_date null if missing.

Narrowing: No tasks without an assigned owner in text. No markdown fences in output. Skip small talk.

Context: [paste transcript]

How RISEN differs from RTF, CO-STAR, and CRAFT

RTF (Role, Task, Format) is the lightest template. Use RTF for quick chat when Steps and End-goal are obvious. RISEN adds explicit sequencing and a done definition. Pick RISEN when teammates ship inconsistent length or miss exclusion rules.

CO-STAR adds Context and Audience as first-class fields. RISEN folds audience into Role and expects Context below the block. Some teams rename Role to "Role and Audience" and keep RISEN headers stable.

CRAFT stresses Context, Role, Action, Format, Tone. RISEN is stronger when you need numbered Steps and a hard End-goal for agent-like tasks without building an agent.

The hub article prompt-frameworks-compared walks all frameworks side by side. This page goes deep on RISEN only so you can train new hires on one skeleton before they browse alternatives.

Step-by-step: build a team RISEN prompt in fifteen minutes

Step 1: Name the recurring job in one line: "weekly VP email," "tier-1 refund reply," "PR security pass."

Step 2: Draft Role from reader and tone, not job title inflation.

Step 3: Write Instructions as one verb-led sentence.

Step 4: Add Steps only where order matters. Cut redundant micro-steps.

Step 5: Define End-goal with format nouns: table, JSON, email, slide outline.

Step 6: List Narrowing rules your legal or brand team already uses.

Step 7: Paste Context last. Run once on GPT-5.6 Sol or Claude Fable 5. Score against End-goal.

Step 8: Store the template in your team wiki. Link https://promptmake.net/frameworks/risen for drafts from rough notes.

Solo builder review pass

Read End-goal first. Does the output match? If not, fix Narrowing before you change Role.

Read Steps. Did the model skip a step? Merge small steps or move critical rules to Narrowing.

Save the winning prompt with date and model name. RISEN templates rot when product names change.

Team handoff pass

Assign one owner per template. Review monthly. Delete examples that embed outdated policies.

New hires fill blanks only for Context on first week. They should not invent new Narrowing rules without review.

Common RISEN mistakes and fixes

Mistake: Role is a long fantasy biography. Fix: three lines max, audience-first.

Mistake: Instructions contain five unrelated tasks. Fix: split prompts or chain with explicit End-goal handoff.

Mistake: Steps repeat Instructions word for word. Fix: milestones only.

Mistake: End-goal says "good summary" without format. Fix: name artifact and length.

Mistake: Empty Narrowing. Fix: add at least length, honesty, and exclusion rules.

Mistake: Context above RISEN so the model misses Steps. Fix: instructions first, paste second.

Mistake: Using RISEN for API structured output without schema. Fix: add JSON schema or use native structured output; RISEN is for chat drafts.

Using PromptMake RISEN generator

Open https://promptmake.net/frameworks/risen. Paste a rough job description. The tool returns labeled Role, Instructions, Steps, End-goal, and Narrowing blocks you can edit.

PromptMake generates text only. It does not execute Steps or send email. Free tiers follow site-wide limits on text paths; confirm quotas on promptmake.net before batch work.

Use the generator to bootstrap templates, then move approved versions to your wiki. Human review still owns Narrowing for regulated industries.

FAQ

What does RISEN stand for in prompting?

RISEN stands for Role, Instructions, Steps, End-goal, and Narrowing. Role sets voice and reader. Instructions state the main task. Steps order the work. End-goal defines the finished artifact. Narrowing lists limits and exclusions. Together they form the risen framework for repeatable chat prompts.

When should I use the risen framework instead of RTF?

Use RISEN when RTF drafts keep missing length limits, skipping sequence, or adding banned content. RTF stays faster for one-shot tasks with obvious format. RISEN earns its five fields on recurring team workflows with review gates.

How is this article different from prompt-frameworks-compared?

Prompt-frameworks-compared is the hub that lines up CO-STAR, RISEN, CRAFT, and RTF for picker decisions. This risen framework page teaches each RISEN field with six full examples and a team workflow. Read the hub to choose; read this page to implement RISEN.

Can I use RISEN with Claude, ChatGPT, and Gemini?

Yes. RISEN is model-agnostic labeled prose. Claude Fable 5, GPT-5.6 Sol, and Gemini 3.5 Flash all follow the five headers when you keep Context after the block. Reasoning-class models still benefit from clear End-goal and Narrowing without chain-of-thought filler.

Where does Context go in a RISEN prompt?

Put Context after the five RISEN fields and before any long pasted source text. Instructions belong first so the model sees Role through Narrowing as the contract, then reads data. Some teams add a Context label line for clarity.

How do I start with RISEN on PromptMake for free?

Open https://promptmake.net/frameworks/risen, describe your job in plain language, and edit the generated five fields. Guest and free account limits apply per tool path on promptmake.net. Register if you hit the guest cap mid-template build.

What belongs in Narrowing versus End-goal?

End-goal describes what success looks like: format, sections, and stopping point. Narrowing describes what to avoid: max length, banned claims, source-only facts, tone limits, and language. If a rule prevents a bad output, it belongs in Narrowing. If it defines the deliverable shape, it belongs in End-goal.