RTF Framework
Three elements. Zero overhead. RTF distills prompting to its absolute essentials — Role, Task, Format — giving you a lightweight structure that works when speed matters more than depth.
Origin: RTF is a community framework from 2023 that emerged as one of the most minimal named prompting structures in widespread use. It addresses a common problem: users who know they should structure their prompts but find multi-element frameworks like CO-STAR or CRISPE intimidating. RTF strips the concept down to three essential questions — who should the AI be (Role), what should it do (Task), and how should the output look (Format). The framework gained traction through social media and blog communities as a beginner-friendly entry point into structured prompting.
Modern LLM Status: RTF remains highly practical as a quick-start framework for everyday prompting. Modern LLMs like Claude, GPT-4, and Gemini respond well to even minimal structure — specifying a role, a clear task, and an output format consistently outperforms unstructured prompts. RTF’s limitation is that it lacks elements for context, constraints, and audience that more comprehensive frameworks provide. For simple to moderate tasks, RTF delivers excellent results with minimal effort. For complex, high-stakes prompts, consider upgrading to a fuller framework like CO-STAR or CRISP.
The Minimum Viable Prompt
Most people write prompts that are either completely unstructured (“Write me a blog post”) or so heavily templated that using the framework takes longer than the task itself. RTF occupies the sweet spot: just enough structure to dramatically improve output quality, without enough overhead to slow you down.
Three questions unlock better results. Role tells the AI what expertise to bring. Task tells it what to do. Format tells it how to present the result. These three elements address the most common failure modes of unstructured prompts: wrong perspective, vague objectives, and unusable output shapes.
Think of RTF like ordering food. Role is telling the chef what cuisine you want. Task is the dish. Format is how you want it served — plated, boxed, or family-style. Three decisions, and you get something far better than “surprise me.”
The persona or expertise the AI should adopt — who it should be for this task.
The specific action or deliverable — what the AI should produce.
The output structure or presentation — how the result should be delivered.
The RTF Process
Three steps from vague request to structured prompt
Role — Assign the Expertise
Tell the AI who it should be. A role activates domain-specific knowledge and vocabulary, shaping the entire response through the lens of that expertise. The more specific the role, the more targeted the output — “senior data analyst at a healthcare company” beats “data person” every time.
“You are an experienced project manager who specializes in agile software delivery for remote teams.”
Task — Define the Action
Specify exactly what the AI should do. A well-defined task includes the action verb, the subject matter, and enough detail to eliminate ambiguity. Avoid compound tasks — if you need multiple deliverables, either chain separate RTF prompts or list the sub-tasks explicitly.
“Create a sprint retrospective template that helps remote teams identify communication bottlenecks and action items for improvement.”
Format — Specify the Output
Tell the AI how to present the result. Format controls everything from structure (bullet list, table, paragraph) to length (250 words, one page) to medium (email, slide deck, code snippet). Without a format directive, the AI defaults to whatever it considers most natural — which may not match what you need.
“Present this as a one-page document with three sections: What Went Well, What Needs Improvement, and Action Items. Use bullet points under each section with space for team members to add their own items.”
See the Difference
How three simple elements transform an unstructured prompt
Unstructured Prompt
Help me with onboarding new employees.
Here are some tips for onboarding: 1) Create a welcome packet. 2) Assign a buddy. 3) Schedule orientation meetings. 4) Set up their workspace. 5) Provide training materials.
RTF Prompt
Role: You are an HR operations manager at a 200-person SaaS company with a remote-first culture.
Task: Create a 30-day onboarding plan for new software engineers that covers tooling setup, team introductions, codebase orientation, and first deliverable expectations.
Format: A week-by-week table with columns for Day, Activity, Owner, and Success Criteria.
A detailed 4-week table with specific daily activities: Week 1 covers environment setup, security training, and buddy pairing. Week 2 focuses on codebase walkthrough and first PR review. Week 3 assigns a starter ticket with mentor support. Week 4 establishes independent workflow with 1:1 check-in cadence. Each row includes the responsible owner and measurable success criteria.
Natural Language Works Too
While structured frameworks and contextual labels are powerful tools, LLMs are exceptionally good at understanding natural language. As long as your prompt contains the actual contextual information needed to create, answer, or deliver the response you’re looking for — the who, what, why, and constraints — the AI can produce complete and accurate results whether you use a formal framework or plain conversational language. But even in 2026, with the best prompts, verifying AI output is always a necessary step.
RTF in Action
See how Role, Task, and Format combine for fast, effective prompts
Role: You are a B2B email marketing specialist who focuses on high open rates and clear calls to action.
Task: Write a follow-up email to prospects who attended our webinar on data security but did not book a demo. The email should reference key takeaways from the webinar and create urgency around a limited-time offer for a free security audit.
Format: A single email under 200 words with a compelling subject line, three short paragraphs, and one clear CTA button text.
Note: Review the generated email for accuracy against your actual webinar content and offer terms before sending. AI-generated marketing copy should always be verified against your brand guidelines.
Subject: The vulnerability you spotted in our webinar? Let’s fix it (free).
Hi [First Name],
Last week you joined 200+ security leaders for our webinar on the three data vulnerabilities most companies miss. The stat that stopped the room: 73% of breaches exploit gaps that a standard audit would catch.
We are opening 15 free security audit slots for webinar attendees this month. Our team will scan your environment for the exact vulnerability patterns we discussed — no commitment, full report yours to keep.
Slots are first-come: 6 of 15 are already claimed.
[Book Your Free Audit]
Role: You are a senior technical writer at a cloud infrastructure company, experienced in writing documentation for developers with varying skill levels.
Task: Write a quickstart guide for setting up a PostgreSQL database on our cloud platform. Cover account creation, database provisioning, connection string configuration, and running a first query.
Format: A numbered step-by-step guide with code blocks for commands, a prerequisites section at the top, and a troubleshooting FAQ with three common issues at the bottom.
Note: Verify all commands and connection strings against your platform’s current documentation before publishing. AI-generated technical instructions should be tested in a development environment first.
A complete quickstart guide structured exactly as requested: Prerequisites (account, CLI tool, PostgreSQL client), then 6 numbered steps with inline code blocks for each command, clearly labeled expected outputs, and a 3-item troubleshooting FAQ covering connection timeouts, authentication errors, and firewall configuration. Each step includes both the command and a one-sentence explanation of what it does.
Role: You are a middle school science teacher with 10 years of experience making complex concepts accessible to 12-14 year olds through hands-on activities.
Task: Design a 45-minute lesson on the water cycle that includes a hands-on demonstration students can do with everyday materials, a class discussion component, and a formative assessment at the end.
Format: A lesson plan with five sections: Learning Objectives (3 bullet points), Materials Needed, Activity Sequence (with time allocations), Discussion Questions (4 questions), and Assessment (a 5-question exit ticket).
Note: Review the lesson plan against your state curriculum standards and adapt the hands-on activity for your specific classroom setup. AI-generated educational content should be verified by qualified educators.
A structured 45-minute lesson plan: Learning Objectives aligned to NGSS standards, a materials list using only items found in a kitchen (pot, ice, plastic wrap, food coloring), a timed activity sequence (5-min intro, 15-min demo, 10-min discussion, 10-min group work, 5-min exit ticket), four Bloom’s Taxonomy-aligned discussion questions progressing from recall to analysis, and a 5-question exit ticket mixing multiple choice with one short-answer explanation.
When to Use RTF
Best for quick, everyday tasks where minimal structure yields maximum improvement
Perfect For
Emails, summaries, outlines, and quick drafts where a lightweight structure is all you need to get a focused, well-formatted response.
RTF is the simplest named framework — perfect for people who are just starting to structure their prompts and want a memorable starting point.
When you need to generate multiple variations quickly, RTF’s low overhead lets you swap roles, tasks, or formats without rewriting the entire prompt.
RTF’s simplicity makes it ideal for building reusable prompt templates where team members fill in their own role, task, and format values.
Skip It When
When audience, tone, and context matter deeply, RTF’s three elements are not enough. Upgrade to CO-STAR or CRISPE for full communication control.
Multi-hop questions, analytical tasks, or problems that require step-by-step reasoning need techniques like Chain-of-Thought or Self-Ask, not structural frameworks.
When you need to specify constraints, evaluation criteria, or examples alongside the core request, RTF’s three slots cannot capture everything. Use CRISP or CRISPE instead.
Use Cases
Where RTF delivers quick wins with minimal setup
Email Drafting
Assign a professional role, specify the email purpose, and define the format (length, tone markers, CTA) for consistent, polished email output.
Content Creation
Blog posts, social media captions, and marketing copy where specifying the writer persona and output format immediately sharpens the voice and structure.
Study Aids
Cast the AI as a tutor in a specific subject, define the concept to explain, and request the output as flashcards, summaries, or practice questions.
Code Generation
Assign a developer role with a language specialty, specify the function or feature, and request the format (commented code, with tests, as a diff).
Interview Prep
Set the AI as a hiring manager in your target industry, ask it to generate interview questions, and specify the format as Q&A pairs with evaluation criteria.
Meeting Preparation
Ask the AI to act as a meeting facilitator, create agendas or talking points, and output them as timed agenda blocks or discussion frameworks.
Where RTF Fits
RTF is the minimalist entry point in the structured prompting spectrum
RTF is intentionally minimal. Once you are comfortable with Role, Task, and Format, you will naturally start wanting to add more context — audience, constraints, tone, examples. That instinct is the signal to graduate to a fuller framework like CRISP, CO-STAR, or CRISPE. Think of RTF not as a ceiling but as a foundation: every comprehensive framework includes RTF’s three elements plus additional dimensions.
Related Techniques & Frameworks
Explore complementary approaches to structured prompting
Start Simple, Start Now
Try RTF on your next prompt or explore fuller frameworks when you need more control.