You don't need 47 prompt engineering techniques. You need a handful of formulas that actually work — and the judgment to pick the right one. Here are seven that hold up across models, tasks, and skill levels.
Most prompt advice falls into two camps: oversimplified tips ("just be specific!") or academic frameworks with more acronyms than a government agency. Neither helps when you're staring at a blank input box trying to get useful output from an AI.
What works is having a small set of proven formulas you can reach for instinctively. Each formula is a different tool for a different job. A wrench isn't better than a screwdriver — it depends on what you're tightening.
We tested these seven formulas across hundreds of real tasks on ChatGPT, Claude, and Gemini. They cover about 95% of what most people need. Learn them, bookmark the templates, and stop reinventing your prompts from scratch every time.
Formula 1: RTCC — Role + Task + Context + Constraints
The workhorse. RTCC is the formula you'll use most often. It works because it answers the four questions every AI model needs: who should I be, what should I do, what do I need to know, and what are the boundaries?
When to use it: General-purpose tasks — writing, analysis, planning, brainstorming. If you're unsure which formula to use, start here.
The structure:
Role: [Who the AI should be]
Task: [Exactly what you need done]
Context: [Background information, audience, situation]
Constraints: [Length, format, tone, things to avoid]Copy-paste templates
Template 1: Content creation
Role: You are a senior content strategist who specializes in B2B SaaS marketing.
Task: Write a LinkedIn post announcing our new feature — automated reporting dashboards.
Context: Our audience is mid-market operations managers who currently build reports manually in spreadsheets. The feature saves approximately 4 hours per week. We launched it yesterday.
Constraints: Keep it under 200 words. Use a conversational but professional tone. Include one specific metric. End with a question to drive engagement. No hashtag spam — 3 maximum.Template 2: Analysis
Role: You are a financial analyst with 15 years of experience in SaaS metrics.
Task: Analyze these quarterly numbers and identify the three most significant trends.
Context: We're a Series B startup with $8M ARR. Last quarter: MRR grew 6%, churn dropped from 4.2% to 3.8%, CAC increased 12%, LTV:CAC ratio is 3.1:1. Board meeting is next week.
Constraints: Structure your analysis with one paragraph per trend. Quantify the impact where possible. Flag anything the board will ask about. Keep it under 400 words.Template 3: Problem-solving
Role: You are a senior DevOps engineer who specializes in AWS infrastructure.
Task: Diagnose why our API latency spiked from 120ms to 800ms this morning and propose a fix.
Context: We're running Node.js on ECS Fargate behind an ALB. No code deployments in the last 48 hours. CloudWatch shows CPU at 45% and memory at 62%. The spike correlates with a 3x increase in traffic from a single IP range.
Constraints: Start with the most likely cause. List diagnostic steps in priority order. Recommend both an immediate fix and a long-term prevention strategy.Write me a LinkedIn post about our new feature.
Role: Senior content strategist for B2B SaaS. Task: LinkedIn post for automated reporting dashboards. Context: Audience is ops managers doing manual spreadsheet reports. Feature saves 4 hours/week. Constraints: Under 200 words, conversational tone, end with a question.
The before prompt could produce anything. The RTCC version gives the model everything it needs to produce something targeted and useful on the first try.
Formula 2: Before/After — State Transformation
The gap closer. Instead of describing what you want from scratch, you show the AI where you are and where you want to be. The model figures out how to bridge the gap.
When to use it: Editing, improving, transforming, or upgrading existing content. Also great for rewriting, tone-shifting, and level-adjusting.
The structure:
Here is my current [thing]:
[paste current version]
Transform it to:
- [Desired quality 1]
- [Desired quality 2]
- [Desired quality 3]
Keep: [what should stay the same]
Change: [what needs to be different]Copy-paste templates
Template 1: Email rewrite
Here is my current email draft:
"Hey, just wanted to check in on the project. Things seem behind schedule and I'm worried about the deadline. Can we talk about this?"
Transform it to:
- Professional but not cold
- Specific about what's behind and by how much
- Includes a concrete proposed next step instead of a vague request to "talk"
Keep: The core concern about timeline
Change: Remove passive language and vague worry. Replace with direct, constructive communication.Template 2: Resume bullet upgrade
Here are my current resume bullets:
- Managed a team of developers
- Worked on the company website
- Helped increase sales
Transform each bullet to:
- Use the XYZ formula (Accomplished X, as measured by Y, by doing Z)
- Include plausible metrics that reflect senior-level impact
- Start with a strong action verb
Keep: The core responsibilities described
Change: Vague descriptions into quantified achievementsTemplate 3: Technical-to-plain-English translation
Here is my current technical explanation:
"The system uses a distributed event-driven architecture with eventual consistency guarantees, leveraging Apache Kafka for message brokering and Redis for caching hot paths."
Transform it to:
- Understandable by a non-technical CEO
- Uses a real-world analogy
- Preserves accuracy — no dumbing down to the point of being wrong
Keep: All the actual technical concepts
Change: Jargon into plain language. Abstract architecture into concrete metaphor.This formula works because AI models are remarkably good at understanding transformations. Showing the gap between "here" and "there" gives the model a clearer target than a description alone.
Formula 3: PAT — Persona + Audience + Task
The empathy formula. PAT forces you to think about who's speaking and who's listening before you think about what to say. That reframing alone produces dramatically better output for any communication task.
When to use it: Anything audience-facing — marketing copy, emails, presentations, educational content, customer communications.
The structure:
Persona: Write as [specific type of person with specific expertise]
Audience: The reader is [specific description of who will read this]
Task: [What to create, with format and length]Copy-paste templates
Template 1: Customer onboarding email
Persona: Write as a friendly but knowledgeable customer success manager at a project management SaaS company. You genuinely want users to succeed, not just retain them.
Audience: The reader is a new user who signed up yesterday. They're a small business owner (5-10 employees) who has never used project management software before. They're probably overwhelmed by the interface.
Task: Write a day-1 onboarding email that gets them to complete one specific action — creating their first project. Keep it under 150 words. One clear CTA. No corporate speak.Template 2: Technical blog post
Persona: Write as a senior developer who explains complex concepts clearly and has strong opinions backed by experience. Think of someone who's shipped production code for 12+ years and teaches on the side.
Audience: Intermediate developers (2-4 years experience) who understand basic concepts but struggle with architecture decisions. They read Hacker News and prefer concrete examples over theory.
Task: Write a 1000-word blog post explaining when to use microservices vs. monoliths. Include at least two real-world scenarios. Take a clear position — don't hedge with "it depends" without following up with specifics.Formula 4: GCO — Goal + Context + Output Format
The efficiency formula. GCO strips prompt engineering down to three essentials. No role assignment, no persona work — just what you need, why you need it, and how you want it delivered.
When to use it: Quick tasks where speed matters more than nuance. Data formatting, list generation, summarization, structured extraction.
The structure:
Goal: [What you need to achieve]
Context: [Why you need it and any relevant background]
Output format: [Exactly how to structure the response]Copy-paste templates
Template 1: Meeting prep
Goal: Prepare me for a meeting with a potential investor in our edtech startup.
Context: The investor focuses on Series A edtech companies. They previously invested in Duolingo and Coursera. Our product is an AI tutoring platform for K-12 math. We have 15,000 MAU and $400K ARR with 20% month-over-month growth.
Output format: A table with three columns: (1) Likely question they'll ask, (2) Recommended answer (2-3 sentences), (3) Data point to cite. Include 8-10 rows, prioritized by likelihood.Template 2: Competitive analysis
Goal: Compare our pricing model against three competitors.
Context: We charge $29/month per user. Competitor A charges $19/month (limited features), Competitor B charges $49/month (enterprise features), Competitor C uses usage-based pricing starting at $0.10 per transaction.
Output format: A comparison table with rows for: price point, target customer, key differentiator, biggest weakness, and one-sentence positioning statement for each. Follow the table with a 3-sentence summary of our competitive advantage.GCO is the formula you use when you don't need the AI to play a character or empathize with an audience. You need information structured and delivered. It's the prompt template equivalent of a well-organized brief.
Formula 5: Chain-of-Thought — Step-by-Step Reasoning
The thinking formula. Instead of asking for an answer, you ask the model to show its reasoning process. This dramatically improves accuracy on any task involving logic, math, analysis, or multi-step decisions.
When to use it: Math, logic, coding, debugging, complex analysis, decision-making, anything where the AI might take a shortcut and get it wrong.
The structure:
[State your question or problem]
Think through this step by step:
1. First, [identify/analyze/consider X]
2. Then, [evaluate/calculate/compare Y]
3. Finally, [synthesize/decide/recommend Z]
Show your reasoning at each step before giving your final answer.For a deep dive into this technique, see our complete guide to Chain-of-Thought prompting.
Copy-paste templates
Template 1: Business decision
Should we hire a full-time content writer or continue using freelancers?
Think through this step by step:
1. First, calculate the annual cost of a full-time hire (salary, benefits, equipment, management time) vs. our current freelancer spend ($4,500/month for 12 articles).
2. Then, evaluate the non-financial factors: quality consistency, institutional knowledge, turnaround time, scalability.
3. Then, consider the risks of each option.
4. Finally, make a recommendation with a clear threshold — at what content volume does the hire become the better option?
Show your reasoning at each step.Template 2: Code debugging
This Python function is returning incorrect results for negative numbers:
def calculate_discount(price, discount_percent):
return price - (price * discount_percent / 100)
When called with calculate_discount(-50, 10), it returns -45 instead of the expected behavior.
Think through this step by step:
1. First, trace the execution with the given inputs.
2. Then, identify what the "expected behavior" should be for negative prices — is this a valid input?
3. Then, consider edge cases: zero price, 100% discount, discount > 100%.
4. Finally, provide a corrected version with input validation and explain each change.Tip
Chain-of-Thought is not just for hard problems. Adding "think step by step" to almost any prompt improves output quality — the model catches its own mistakes when forced to show its work.
Formula 6: Few-Shot — Learning from Examples
The pattern formula. Instead of explaining what you want in words, you show the AI examples of inputs and outputs. The model learns the pattern and applies it to new inputs.
When to use it: Consistent formatting, classification, data transformation, style matching — any task where "I'll know it when I see it" is easier than writing a specification.
For the full breakdown, see our few-shot prompting guide.
The structure:
[Brief task description]
Examples:
Input: [example 1 input]
Output: [example 1 output]
Input: [example 2 input]
Output: [example 2 output]
Input: [example 3 input]
Output: [example 3 output]
Now apply the same pattern:
Input: [your actual input]
Output:Copy-paste templates
Template 1: Product description style matching
Write product descriptions matching this style and format.
Examples:
Input: Wireless Bluetooth headphones, 40-hour battery, noise canceling
Output: Block out the world (for 40 hours straight). These wireless headphones deliver studio-quality noise cancellation that makes open offices bearable and long flights pleasant. One charge. One full work week.
Input: Standing desk converter, adjustable height, fits on existing desk
Output: Your desk, upgraded. This converter transforms any table into a sit-stand workspace in 30 seconds. No tools, no new furniture, no excuses to sit all day.
Now apply the same pattern:
Input: Mechanical keyboard, hot-swappable switches, RGB backlight, compact 65% layout
Output:Template 2: Customer feedback classification
Classify customer feedback into categories and sentiment.
Examples:
Input: "The app crashes every time I try to export a PDF"
Output: Category: Bug Report | Sentiment: Negative | Priority: High | Feature: Export
Input: "Love the new dashboard! Much easier to find my reports now"
Output: Category: Positive Feedback | Sentiment: Positive | Priority: Low | Feature: Dashboard
Input: "Would be nice if you could integrate with Slack"
Output: Category: Feature Request | Sentiment: Neutral | Priority: Medium | Feature: Integrations
Now apply the same pattern:
Input: "I've been waiting 3 days for a response from support"
Output:Few-Shot is the formula that feels like magic the first time you use it. The AI picks up on patterns you'd struggle to articulate — tone, structure, level of detail, formatting quirks — all from examples.
Formula 7: Iterative Refinement — Generate Then Improve
The perfectionist's formula. Instead of trying to get a perfect result in one shot, you explicitly build revision into the process. Generate a first draft, evaluate it against criteria, then improve.
When to use it: High-stakes content (sales pages, presentations, important emails), creative work, anything where "good enough" isn't good enough.
The structure:
Step 1: [Generate initial output with basic instructions]
Step 2: Now review what you just wrote against these criteria:
- [Criterion 1]
- [Criterion 2]
- [Criterion 3]
Score each criterion 1-10 and explain gaps.
Step 3: Rewrite to address every gap you identified. The final version should score 8+ on every criterion.Copy-paste templates
Template 1: Sales email refinement
Step 1: Write a cold outreach email to a VP of Marketing at a D2C skincare brand. We sell an AI-powered social media scheduling tool. Our angle: we can predict which posts will perform best before they publish.
Step 2: Now review this email against these criteria:
- Opens with something relevant to THEM, not about us (1-10)
- Includes one specific, credible proof point (1-10)
- CTA is low-friction — not "book a demo" but something easier (1-10)
- Under 100 words (1-10)
- Would pass the "would I read this?" test from a busy VP (1-10)
Score each and identify the biggest weakness.
Step 3: Rewrite the email to score 8+ on every criterion. Show the final version only.Template 2: Presentation slide content
Step 1: Write the content for a 10-slide investor pitch deck for an AI-powered hiring platform. We reduce time-to-hire by 60% and have $2M ARR with 150% net revenue retention.
Step 2: Review each slide against these criteria:
- One core idea per slide (no information overload)
- Data points are specific and compelling
- Narrative builds logically from problem to solution to traction to ask
- No buzzword bingo — every claim is backed by a number or example
Identify the two weakest slides.
Step 3: Rewrite those two slides to match the quality of the strongest ones.Iterative Refinement works because it mirrors how human experts actually work. Nobody writes a perfect first draft. But most people prompt AI once and accept whatever comes back. Building evaluation into the prompt changes that.
The Decision Matrix: Which Formula to Use
Choosing the right formula shouldn't require a decision. Here's the quick reference:
| Situation | Formula | Why |
|---|---|---|
| General task, not sure where to start | RTCC | Covers all bases, works for everything |
| Improving or transforming existing content | Before/After | Easier to show the gap than describe the destination |
| Writing for a specific audience | PAT | Forces audience empathy into the output |
| Quick structured output, no frills | GCO | Fastest path from need to result |
| Logic, math, analysis, or debugging | Chain-of-Thought | Prevents the model from jumping to wrong conclusions |
| Need consistency, or "match this style" | Few-Shot | Shows don't tell — examples beat specifications |
| High-stakes, needs to be excellent | Iterative Refinement | Built-in quality control catches what one-shot misses |
Start with RTCC if you're unsure — it handles 80% of tasks well enough.
Switch to a specialist formula when RTCC doesn't quite fit — audience-heavy tasks get PAT, transformation tasks get Before/After, analytical tasks get Chain-of-Thought.
Combine formulas for complex work — use RTCC for the initial prompt, then Iterative Refinement to polish the output. Or use Few-Shot examples inside an RTCC structure.
Stop Memorizing, Start Building
You don't need to memorize these formulas. You need to use them until they become instinct. Bookmark this page, copy the templates, and modify them for your actual tasks.
Or skip the manual work entirely. SurePrompts' AI Prompt Generator applies these formula principles automatically — describe what you need in plain English and it builds a structured, effective prompt for you. The Template Builder has 320+ templates that already bake these patterns in.
The difference between mediocre AI output and excellent AI output isn't the model. It's the formula. Pick the right one, fill in the blanks, and watch the quality gap disappear.
For a broader look at prompt frameworks including CRAFT, RACE, and RISEN, see our complete framework comparison. And if you're just getting started, the prompt engineering basics guide covers the fundamentals you'll build on.