How to Write AI Prompts That Actually Work
Master the skill of communicating with AI. Learn our proven PCRF framework, see real examples, and start getting better results from ChatGPT, Claude, and Gemini.
Last updated: September 5, 2026
What is an AI prompt?
An AI prompt is a text instruction that tells an AI system like ChatGPT, Claude, or Gemini what you want it to do. To write an effective AI prompt: assign the AI a persona, provide context, make a specific request, and define the output format (the PCRF framework). The quality of your prompt directly determines the quality of the response — specific, structured prompts consistently outperform vague ones. For the broader discipline these techniques belong to, see what is prompt engineering.
How to Write a Prompt in 6 Steps
To write a prompt, open with the role you want the AI to take, add the context it cannot infer, state one specific request, define the output format, add constraints such as length and tone, then read the result and refine the weakest instruction. Six steps, in that order.
-
1
Start with a clear role
The role sets vocabulary, depth and default assumptions before the model reads the task. A prompt that opens with "You are a senior financial analyst" gets a different first draft than the same request with no role at all, because the model has been told which conventions to apply.
You are a senior financial analyst who briefs executives.
-
2
Provide necessary context
Most weak outputs are missing context, not missing intelligence. Name the audience, the situation, the decisions already made and the material to use. If the answer depends on a document, paste the document rather than describing it.
Context: the reader is a VP of Operations at a mid-size manufacturer who has already seen the demo.
-
3
Make one specific request
Draft, compare, summarize, rewrite, classify. One verb, one deliverable. When a task genuinely has several parts, number them inside the request so nothing is dropped, or split it across separate prompts and carry the good output forward.
Task: draft a 150-word follow-up email that references their inventory question.
-
4
Specify the output format
Format instructions are the cheapest quality gain available. Asking for "a four-row table with columns Risk, Likelihood, Owner, Mitigation" removes an entire round of editing compared with asking for "an overview of the risks".
Format: a four-row table with the columns Risk, Likelihood, Owner, Mitigation.
-
5
Add constraints and exclusions
Constraints are how you keep a draft usable. Set a word ceiling, name the words or claims to avoid, and tell the model what to do when it does not know something rather than letting it fill the gap.
Constraints: under 200 words, no jargon, and say "not in the source" for anything the document does not state.
-
6
Read the output and refine
Iteration beats restarting. Identify the one element that produced the gap, whether it was the role, the missing context or a vague format, and change that line only. When a prompt works, save it so the next version starts from the working one.
Keep the structure, but rewrite the second paragraph in plain language and cut it to two sentences.
The six steps are the same four elements of the PCRF framework below, plus the two habits that separate a usable first draft from a second attempt: constraints, and refining one line at a time.
Copy-Paste AI Prompt Templates
Five templates that work as written in ChatGPT, Claude, Gemini, and any other chat interface. Copy one, replace the text in square brackets, and delete any line the task does not need.
Any task. Fill the five slots and delete the ones a simple task does not need.
Follow-ups, internal updates, and anything that has to land in one read.
Reports, transcripts, contracts, and anything you have to brief someone else on.
Turning rows of numbers into something a decision can be made from.
Tightening something you have already drafted without losing your voice.
Keep the versions that work for your role in one place. A prompt you have already tuned is worth more than a new one, and reusing it is what makes the output consistent from week to week.
The PCRF Framework for Better Prompts
After analyzing thousands of prompts, we developed the PCRF framework—a simple structure that dramatically improves AI outputs. Every effective prompt includes these four elements:
PCRF: The 4 Elements of Effective Prompts
Not every prompt needs all four elements—simple tasks can work with just Request and Format. But complex tasks benefit significantly from the full framework.
To go past self-teaching, see our review of the best prompt engineering courses.
100 AI prompts every professional should steal.
- Categorized by job function — find what you need in seconds
- Each prompt includes a real use case and constraint structure
- Works with ChatGPT, Claude, Gemini, and Copilot
Instant download. We'll also email you a copy. No spam.
Example 1: Content Creation
Let's see the PCRF framework in action. Compare these two prompts for writing a blog post:
The weak prompt is vague and will produce generic content. The strong prompt specifies the persona (B2B marketing strategist), context (SaaS companies), request (500-word blog post with 3 tactics), and format (headers and bullet points).
12 Tips for Writing Better AI Prompts
These techniques work across ChatGPT, Claude, Gemini, and other AI platforms. Bookmark this list and reference it when crafting prompts.
Be Specific, Not Vague
Replace "Write something about marketing" with "Write a 500-word guide to email subject lines for B2B software companies." Specificity eliminates ambiguity.
Assign a Relevant Role
Start with "You are an experienced [role]..." This primes the AI to respond from that perspective, improving accuracy and relevance.
Provide Context
Include background information: who is the audience, what is the situation, what constraints exist. The more context, the better the output.
Specify Output Format
Tell the AI exactly how to structure the response: "Format as a numbered list," "Use headers and bullet points," "Write as a table with columns for X, Y, Z."
Set Length Constraints
Be explicit about length: "in 100 words," "in 3 paragraphs," "as a one-page summary." Without constraints, AI often over-produces — and every extra word is billed output, so tight length limits are a simple way to cut AI token costs.
Give Examples
Show the AI what good output looks like. "Here's an example of the tone I want: [example]." This technique (few-shot prompting) dramatically improves consistency.
Ask for Multiple Options
Instead of "Write a headline," try "Write 10 headline options." This gives you choices and reveals the AI's range of ideas.
Use "Step by Step"
For complex reasoning, add "Think through this step by step." This chain-of-thought technique improves accuracy on analytical tasks.
Specify What to Avoid
Include negative instructions: "Do not use jargon," "Avoid clichés like 'game-changer,'" "Don't include a conclusion section."
Define the Tone
Be explicit about voice: "professional but conversational," "formal and authoritative," "friendly and approachable." Tone shapes the entire output.
Break Complex Tasks into Steps
Instead of one massive prompt, use sequential prompts: first outline, then expand each section. This maintains quality throughout.
Iterate and Refine
Your first prompt rarely produces perfect output. Review, identify what's missing, and refine. Prompt engineering is an iterative process.
Example 2: Email Writing
Emails are one of the most common AI use cases. Here's how specificity transforms results:
The strong prompt includes context about the recipient, the situation, their specific interests, and the desired tone. This level of detail produces emails that feel personalized rather than generic.
Common Prompting Mistakes to Avoid
- Vague requests without specifics
- Missing context or background
- No format specification
- Asking for too much at once
- Accepting first output without iteration
- Using AI jargon the model doesn't need
- Be specific about what you want
- Provide all relevant background
- Specify exact output format
- Break complex tasks into steps
- Refine prompts based on output
- Write naturally, as if to a colleague
Example 3: Document Summarization
Summarizing long documents is another common use case. Compare these approaches:
The strong prompt specifies the audience (CEO), what to focus on (four specific areas), the format (one-page brief with bullets), and the style (plain language). This produces an immediately useful summary rather than a generic overview.
The Key Insight: Practice Beats Theory
Reading about prompting is not the same as doing it. Like any skill, prompt engineering improves through practice and feedback. Research shows that employees with formal training achieve 2.7x higher proficiency than self-taught users.
The most effective way to improve is through structured practice where you:
- Write prompts for real work tasks (not hypothetical exercises)
- Receive feedback on what could be improved
- Iterate based on that feedback
- Build a library of prompts that work for your role
Where to Take an AI Prompt Writing Course
A guide gets you to a good first draft. What moves the skill is writing prompts for your own work and getting told what is weak about them. The Iternal AI Academy scores and critiques prompts as you write them, so the feedback arrives on your task rather than on a sample exercise, and the technique transfers to ChatGPT, Claude, Gemini, and any tool your team adopts later.
Rolling this out to a whole department is a different problem from learning it yourself. For that, see AI training for employees.
Frequently Asked Questions
Tools without training stall. Train the team that runs it.
Master Prompt Engineering with Hands-On Practice
Iternal AI Academy offers 810+ courses with interactive AI exercises that score and critique your prompts in real-time. Build skills that transfer to ChatGPT, Claude, Gemini, and any AI tool.
Start Learning