The RTFC Framework: How to Write AI Prompts That Actually Work
The Problem with Most Prompts
90% of people use AI like this:
"Write me a blog post about marketing."
And they get generic, boring, surface-level output. Then they blame the AI.
The problem isn't the model. The problem is the prompt.
The RTFC Framework
After testing 500+ prompts across ChatGPT, Claude, and Gemini, I've found that every great prompt has 4 elements:
R Role — Tell the AI who it is. This primes it with the right knowledge and tone.
✗ "Write a marketing email"
✓ "You are a senior email marketing copywriter who has
written campaigns for Nike, Apple, and Spotify."
T Task — Be specific about what you want. Not "write content" — what kind? How long? For whom?
✗ "Write content for my landing page"
✓ "Write a landing page hero section (headline +
subheadline + CTA) for a B2B AI tool targeting startup founders."
F Format — Specify the output structure. This eliminates 90% of editing.
✗ "Give me content ideas"
✓ "Give me 10 content ideas in a markdown table with
columns: Idea, Hook, Platform, Estimated Time, Difficulty (1-5)"
C Constraints — Set rules and boundaries. This is where most people go wrong by being too permissive.
✗ "Write a blog post"
✓ "Write a 1,200-word blog post. Conversational tone.
Include 3 real-world examples. No jargon. Reading level: 8th grade.
Include a TL;DR at the top."
Putting It All Together
Here's a complete RTFC prompt:
Role: You are a senior sales strategist who has helped
50+ B2B startups increase their demo booking rate by 30%+.
Task: Write a cold email outreach sequence (3 emails)
for a SaaS tool that helps agencies automate client reporting.
Format: For each email, provide:
- Subject line (with A/B variant)
- Email body (under 150 words)
- Send timing (Day 1, Day 3, Day 7)
- Single CTA
Constraints:
- Tone: confident but not pushy
- Include one specific pain point per email
- No buzzwords ("synergy", "revolutionary")
- Each email must work standalone
- Personalization placeholders in [brackets]
The output from this prompt is immediately usable. No editing needed.
5 More Prompts Using RTFC
1. Content Calendar Generator
Role: Senior content strategist, 10 years in B2B SaaS.
Task: Create a 30-day content calendar.
Format: Table with Day, Topic, Platform, Hook, CTA.
Constraints: Mix 70% educational, 30% promo. 2 SEO keywords per post.
2. Customer Research Synthesizer
Role: UX researcher with expertise in qualitative analysis.
Task: Analyze 20 customer interview transcripts, identify top 5 pain points.
Format: Numbered list with: Pain Point, Frequency, Severity (1-5), Quote.
Constraints: Prioritize by business impact. Include original quotes.
3. Competitive Analysis
Role: Strategy consultant at McKinsey.
Task: Compare 3 competitors in the AI prompt tool space.
Format: Matrix table: Pricing, Features, Target Market, Strengths, Weaknesses.
Constraints: Be objective. Use data, not opinions. End with recommendation.
4. Code Reviewer
Role: Senior software engineer, Python, 15 years experience.
Task: Review this function for bugs, performance, and readability.
Format: Line-by-line comments + summary score (1-10) + refactored version.
Constraints: Focus on production readiness. Flag security issues with ⚠️.
5. Social Media Repurposer
Role: Social media manager who grew 3 accounts to 100K+ followers.
Task: Turn this blog post into 5 tweets, 1 LinkedIn post, and 1 thread outline.
Format: Each piece ready to copy-paste. Include hashtags.
Constraints: Adapt tone per platform. Max 280 chars per tweet.
Why This Works
AI models are prediction engines. They predict the most likely next word based on context. When you give vague prompts, the AI falls back to its training average — which is mediocre.
When you specify role, task, format, and constraints, you constrain the prediction space. The output becomes specific, structured, and immediately useful.
Think of it like ordering at a restaurant:
- ✗ "Give me food" → You get whatever they feel like making
- ✓ "I'll have the ribeye, medium rare, with asparagus, no butter" → You get exactly what you want
Get 500+ More Prompts
The full toolkit has 200+ production-ready prompts using RTFC across 15 categories.
Free Sample on GitHub Browse Premium PacksUse coupon GITHUB20 for 20% off.
FAQ: RTFC Framework
What does RTFC stand for?
RTFC stands for Role, Task, Format, Constraints. Some practitioners use "Criteria" for the C instead of "Constraints" — both refer to the same concept: defining quality standards and rules the AI must follow.
RTFC Constraints vs Criteria — what's the difference?
Practically, nothing. "Constraints" implies rules the AI must not break (negative constraints like "no jargon, max 150 words"). "Criteria" implies quality standards the output should meet (positive criteria like "persuasive tone, professional"). We recommend using both: tell the AI what to do AND what not to do.
Is RTFC better than CRISPE or RACE?
For production and business use, yes. RTFC produces structured, consistent output that requires less editing. See our full comparison of RTFC vs CRISPE vs RACE →
Does RTFC work with all AI models?
Yes. RTFC is model-agnostic. It works with ChatGPT (GPT-4, GPT-4o), Claude (3.5 Sonnet, Opus), Gemini (Pro, Ultra), and open-source models like Llama and Mistral.
Where can I learn more about prompt engineering?
Check out our 50 AI Prompts for Business guide, our AI Agent Prompt Engineering deep dive, or browse the free templates on GitHub.