Mastering AI Prompting with CRISPE

Learn how to transform vague AI instructions into clear, professional, high-performing prompts using a 5-step structured framework.

Why Use a Prompting Framework?

AI models excel when given precise context, roles, and constraints. CRISPE helps you organize your thoughts into a clear checklist so you get the exact response you want on the first try.

Naive Prompting

“Write an email to my team about project delays.” (Vague context, unpredictable tone, generic results)

CRISPE Prompting

Defines exact role, background details, clear task goals, target tone, and asks for alternative options to select from.

The 5 Pillars of CRISPE

Step 1 — C/R

Capacity & Role

The Concept: Assigning the AI a specific job title, mindset, and background experience.
How It Works: Tell the model who it should pretend to be. This primes the AI to draw upon domain-specific knowledge and communication patterns.
Real-World Example:

Prompt Snippet: “Act as a Senior Operations Director with 15 years of experience leading cross-functional tech teams.”

Step 2 — I

Insight

The Concept: Providing necessary background facts, context, and operational environment.
How It Works: Give the AI key details like team size, current challenges, budget limits, or target audiences to prevent assumptions or generic advice.
Real-World Example:

Prompt Snippet: “Background: Our software migration is 2 weeks behind schedule due to unexpected security auditing delays. Team morale is slightly stressed.”

Step 3 — S

Statement

The Concept: The core action item or non-negotiable directive you need the AI to complete.
How It Works: Use clear action verbs to define the primary task, concise goals, and required output structure.
Real-World Example:

Prompt Snippet: “Draft a 200-word status update email to stakeholders explaining the delay, revised milestone dates, and key mitigation steps.”

Step 4 — P

Personality

The Concept: Controlling the style, tone, voice, and formatting of the output.
How It Works: Guide how the AI presents information—whether transparent, empathetic, formal, bulleted, or conversational.
Real-World Example:

Prompt Snippet: “Tone: Empathetic, calm, highly professional, and reassuring. Avoid technical jargon or defensive language.”

Step 5 — E

Experiment

The Concept: Requesting multiple options or creative variations to compare and select the best version.
How It Works: Ask the AI for 2–3 different variations (e.g., direct vs. gentle) so you have immediate choices.
Real-World Example:

Prompt Snippet: “Provide 2 versions: Version A focused on timeline adjustments, and Version B focused on team risk mitigation.”

Sources & Reference Reading: Praxis Library - CRISPE Guide | OpenAI Prompt Engineering Documentation

AI Focus practice tool

Prompt Polisher

Type a rough prompt. The tool checks it against the five CRISPE pillars, fills in what is missing, and builds a polished version you can copy.

Your rough prompt

Tip: Ctrl+Enter also polishes. Nothing you type leaves your browser.

Your prompt0/5

Polished prompt0/5

Before and after

Your original

Polished with CRISPE

You can edit this box before copying.

Let the AI ask me questions first

AI Focus Learning Hub

Mastering Prompt Engineering

Learn how to craft powerful instructions for modern AI models—from basic conversational questioning to structured task validation and automated tool calling.

The 4 Golden Rules of Effective Prompting

Before diving into advanced workflows, master these essential foundational rules for consistent AI outputs.

Assign a Persona

Define clear roles. Telling an AI to act as a “Senior Marketing Specialist” or “Python Developer” anchors its tone and vocabulary.

Provide Context

Avoid isolated questions. Supply relevant background details, target audience, and business goals to eliminate guesswork.

Set Constraints

Explicitly detail what NOT to include, length limits, and strict formatting requirements (e.g., bullet points, JSON, plain text).

Chain-of-Thought

Ask the AI to “think step-by-step” before delivering its final answer. Reasoning paths significantly reduce logical errors.

1. Chat-bot Questioning

Compare how vague prompts lead to generic answers, while structured questioning yields actionable insights.

Weak Prompt

“Give me advice on starting a coffee shop.”

Generic AI Output

Starting a coffee shop requires a business plan, finding a good location, buying espresso machines, hiring baristas, and marketing your grand opening.

Mastered Prompt

“Act as a retail consultant. Provide a 3-step checklist for a specialty coffee shop targeting remote workers. Include estimated budget ranges per step in bullet points.”

Structured AI Output

1. Location & Infrastructure ($3k-$8k): 1,200 sq ft unit with high-speed fiber internet.
2. Equipment ($15k-$25k): Lease a dual-boiler espresso machine.
3. Workspace Layout ($5k-$10k): Install quiet zones and ergonomic seating.

Foco’s Pro-Tip

Always specify the desired format and target audience. This narrows down the model’s vast knowledge base to exactly what you need!

2. Task Execution & Validation

When using AI for programmatic tasks, enforce strict validation structures like JSON schema to ensure zero parsing errors.

"prompt_rule": "Extract user sentiment and keywords from feedback.", "required_output": { "sentiment": "POSITIVE | NEGATIVE | NEUTRAL", "confidence_score": 0.98, "tags": ["customer-support", "speed"] }

3. Complex Tasks & Tool Calling

Modern AI agents don’t just generate text—they select dynamic tools (APIs, calculators, web search) to perform complex live tasks.

1. User Request

“What is the weather in Tokyo and calculate $450 × 12?”

2. Model Intent

AI detects tools needed and formats parameters.

Tool Detector

3. API Execution

Executes get_weather() & calculator() live.

API Call

4. Final Synthesis

“Tokyo is 18°C & sunny. The total is $5,400.”

Next step

Watch it explained

Short videos walk through the same ideas in plain English.

Watch the videos →