Prompt Formulas: 12 Proven Patterns for AI Prompting

Situation → Task → Constraints → Output. Our flagship framework for complex, multi-constraint prompts. Provides the model with full context, clear instructions, explicit boundaries, and a defined output format.

Situation: I run a B2B SaaS startup with 200 customers.
Task: Write a churn-prevention email for users inactive 14+ days.
Constraints: Under 150 words, empathetic tone, include one specific feature benefit.
Output: Email with subject line, body, and CTA button text.

Best for: Complex business tasks with multiple requirements, client-facing content, multi-constraint generation.

Instructs the model to break down reasoning into explicit sequential steps before reaching a conclusion. Dramatically improves accuracy on math, logic, and multi-step problems.

Solve this step by step:
A store sells notebooks for £3.50 each. If I buy 7 notebooks and pay with a £50 note, how much change do I receive? Think through each step.

Best for: Math problems, logical reasoning, multi-step analysis, debugging, decision-making.

Provides 2-5 input/output examples before the actual task. The model learns the desired format, tone, and pattern from examples rather than explicit instructions.

Classify the sentiment:

Input: "The product arrived on time and works perfectly" → Positive
Input: "Took 3 weeks to ship and the box was damaged" → Negative
Input: "It's okay, nothing special" → Neutral

Input: "Best purchase I've made this year, exceeded all expectations" →

Best for: Classification, formatting, tone matching, data transformation, pattern replication.

Assigns the model a specific expert identity with domain knowledge, communication style, and professional context. Significantly improves domain-specific output quality.

You are a senior DevOps engineer with 15 years of experience in Kubernetes, AWS, and CI/CD pipelines. You give concise, practical advice and always consider security implications.

Question: How should I handle secrets in my GitHub Actions workflows?

Best for: Domain expertise, consistent voice, professional advice, educational content.

Extends Chain-of-Thought by exploring multiple reasoning paths simultaneously, evaluating each branch, and selecting the strongest. Best for problems with multiple valid approaches.

Consider three different approaches to this problem. For each approach:
1. Describe the strategy
2. Work through the solution
3. Identify strengths and weaknesses

Then select the best approach and explain why.

Problem: How should a 50-person startup prioritise between shipping features and paying down tech debt?

Best for: Strategic decisions, complex problem-solving, creative exploration, architectural choices.

The model first generates an outline (skeleton) of the answer, then fills in each section. Produces more coherent long-form content by planning structure before detail.

First, create a brief outline with 5-7 key sections for an article about prompt engineering best practices. Then expand each section into 2-3 detailed paragraphs.

Best for: Long-form content, articles, reports, documentation, comprehensive guides.

Reasoning + Acting. The model alternates between thinking (reasoning about the task) and acting (calling tools, searching, executing code). The foundation of modern AI agent architectures.

You have access to: search(query), calculate(expression), lookup(term).

Thought: I need to find the current population of Tokyo.
Action: search("Tokyo population 2026")
Observation: 13.96 million
Thought: Now I can answer the question.
Answer: Tokyo's population is approximately 13.96 million.

Best for: AI agents, tool-augmented tasks, research workflows, multi-step data gathering.

The model generates an initial response, then critiques its own output, identifies weaknesses, and produces an improved version. Self-correction without human feedback.

Write a product description for noise-cancelling headphones.

Then review your description critically: Is it specific enough? Does it highlight unique benefits? Is the tone appropriate for a premium brand? Rewrite based on your critique.

Best for: Content refinement, quality improvement, self-editing, catching errors, iterative polish.

Using an LLM to generate or optimise prompts for another LLM (or itself). The "prompt that writes prompts". Powerful for prompt automation and scaling.

I need a prompt that will make an AI generate high-quality product descriptions for an e-commerce store. The descriptions should be 50-100 words, highlight 3 key features, include a benefit statement, and end with a call to action. Write the optimal prompt.

Best for: Prompt automation, template creation, prompt library building, scaling prompt engineering.

Explicitly defines output format constraints — JSON schema, XML structure, specific fields, character limits. Forces the model to produce machine-parseable, structured output.

Extract the following fields from this customer email and return ONLY valid JSON:
{
  "sentiment": "positive|negative|neutral",
  "urgency": "low|medium|high",
  "category": "billing|technical|general",
  "summary": "one sentence"
}

Best for: API responses, data extraction, structured outputs, pipeline integration, JSON/XML generation.

Multi-turn approach where each prompt builds on the previous output, progressively refining towards the desired result. Mimics human editing workflow.

Turn 1: Write a rough draft of an executive summary for our Q1 results.
Turn 2: Make the tone more confident and add specific revenue figures.
Turn 3: Shorten to 150 words and add a forward-looking closing statement.

Best for: Complex content creation, progressive detail addition, collaborative writing, document drafting.

Multiple personas review the same output sequentially — e.g., a writer drafts, an editor refines, a fact-checker validates. Simulates a multi-person workflow in a single prompt.

Step 1 (Writer): Draft a blog post introduction about AI in healthcare.
Step 2 (Editor): Review for clarity, remove jargon, improve flow.
Step 3 (Fact-Checker): Flag any claims that need citations or could be inaccurate.
Step 4 (SEO Specialist): Optimise for the keyword "AI healthcare applications".

Best for: Content production pipelines, quality assurance, multi-perspective review, complex editorial workflows.

Few-shot extraction minimizes context window usage vs zero-shot verbose.

3 well-crafted few-shot examples (150 tokens) outperform a 600-token verbose instruction block, saving 75% on input costs per request.

Without concise few-shot examples, developers write lengthy prose instructions that consume 4x more tokens for equivalent or inferior output quality.

JSON Schema enforcement eliminates parse errors.

OpenAI structured outputs with JSON Schema achieve 99.9% schema adherence vs <70% with unconstrained generation — a 30x reduction in parse failures.

Without schema enforcement, every 1M requests generate 300K+ malformed responses requiring retries, error handling, and downstream data corruption.

Chain-of-thought prompting improves complex reasoning accuracy.

Adding 'Let's think step by step' improves accuracy on GSM8K math benchmarks from 17.7% to 78.7% — a 4.4x improvement on multi-step reasoning tasks.

Without chain-of-thought, models attempt to produce answers in a single leap, failing on problems requiring intermediate steps.