Back to blog
10 min read

How to Write AI Prompts (Step-by-Step Guide)

Learn how to write AI prompts that actually work. A practical, beginner-friendly guide with examples, a simple framework, and common mistakes to avoid.

Writing a good AI prompt means giving the model a clear role, a specific task, relevant background, and a desired format - so it can produce exactly what you need on the first or second try. Use the role → task → context → format framework and you will cover the vast majority of situations you will ever encounter.


TL;DR


Why Your Prompts Probably Aren't Working Yet

Most people type something short - "Write me a bio" or "Explain this contract" - and then feel disappointed when the answer is generic. That is not an AI problem. It is an input problem.

AI models are pattern-completion engines. They produce the most statistically reasonable response to whatever you give them. A vague input produces a vague output, every time. The fix is not a smarter tool - it is a better-structured prompt.

The good news: once you understand the structure, writing effective prompts becomes second nature. It is less like learning to code and more like learning to give clear instructions to a new colleague.


The Core Framework: Role → Task → Context → Format

This four-part framework works for how to write AI prompts step by step, regardless of which tool you use.

1. Role - Tell the AI who to be

Start by giving the AI a persona or area of expertise. This shapes the vocabulary, tone, and assumptions it brings to the answer.

2. Task - Say exactly what you want

Be specific about the action. Use a strong verb: write, summarise, rewrite, compare, list, translate, explain.

3. Context - Give it the background it needs

Context is the most commonly skipped step, and skipping it is the main reason outputs feel off. Tell the AI who the audience is, what the purpose is, and any constraints it should know about.

4. Format - Specify how you want the answer

Do you want bullet points or prose? A table? A short answer or a detailed one? If you do not say, the AI will choose - and it may choose a format that does not fit your workflow.


Putting It Together: Prompts With Examples

Here is how the framework looks in practice. These examples show how to write AI prompts with examples that move from weak to strong.

Weak prompt:

"Write a product description."

Strong prompt:

"Act as an e-commerce copywriter. Write a product description for a reusable water bottle aimed at commuters who care about sustainability. Keep it under eighty words, use a friendly tone, and end with a one-sentence call to action."

Notice that the strong version specifies the role (e-commerce copywriter), the task (write a product description), the context (reusable bottle, commuters, sustainability focus), and the format (under eighty words, friendly tone, CTA at the end).

Another example - summarising a document:

Weak prompt:

"Summarise this."

Strong prompt:

"Act as a business analyst. Summarise the following meeting notes for an executive who was not present. Pull out the three most important decisions made and any action items with owners. Use bullet points."

For more worked examples across different use cases, see Prompt engineering examples for beginners.


How to Write AI Prompts for Beginners: Start Small and Iterate

If you are new to this, the best approach is to start with a real task you already need to do - not a practice exercise. Use the framework, run the prompt, and then read the output critically.

Ask yourself: What did the AI get wrong, and what information was it missing? Then add that information to your next prompt. This iteration loop - prompt, evaluate, refine - is how experienced users actually work.

You do not need to write the perfect prompt on the first try. Think of it as a short conversation, not a single command.


As AI Tools Improve, the Fundamentals Still Win

AI assistants have become more capable over time, and that is genuinely useful. But more capable does not mean mind-reading. Even the most advanced models produce better results when you give them a clear role, a specific task, relevant context, and a preferred format.

The fundamentals of good prompting - specificity, structure, and iteration - are not a workaround for weak AI. They are the baseline for getting strong results from any AI, at any capability level. Investing in these habits now means every future improvement in the tools benefits you more, not less.

If you want to understand the deeper mechanics of why structure matters so much, what is prompt engineering is a useful next read.


Common Mistakes (and How to Fix Them)

Mistake 1: Being too vague

The problem: "Write something about my business" gives the AI almost nothing to work with.

The fix: Use the role-task-context-format framework. Every element you add narrows the solution space and improves the output.


Mistake 2: Asking for too many things at once

The problem: "Write a blog post, create five social media captions, suggest a title, and give me an SEO meta description" in a single prompt usually produces mediocre results across all four tasks.

The fix: Break complex work into a sequence of focused prompts. Ask for an outline first, then expand each section, then refine the tone.

For example:

Prompt 1: "Act as a content strategist. Give me a five-point outline for a blog post about home composting for beginners. Use bullet points."

Prompt 2: "Now expand the second bullet point - 'What you can and cannot compost' - into two short paragraphs written for a general audience."

Each prompt builds on the last, and you stay in control of the direction at every step.


Mistake 3: Forgetting to specify a format

The problem: You wanted a quick bullet-point summary and got five dense paragraphs.

The fix: Always end your prompt with a format instruction. "Answer in bullet points", "Keep it under three sentences", "Use a table with two columns" - these small additions save significant editing time.


Mistake 4: Not giving the AI a role

The problem: Without a role, the AI defaults to a generic helpful assistant, which is fine for simple queries but limiting for anything nuanced.

The fix: Match the role to the task. Legal explainer, marketing copywriter, patient tutor, senior developer - the right role shifts the register and depth of the response meaningfully.


Mistake 5: Giving up after one bad response

The problem: A poor first output leads people to conclude that AI "doesn't work" for their task.

The fix: Treat the first response as a rough draft. Reply with specific feedback: "This is too formal - rewrite it in a conversational tone" or "The second paragraph is off-topic, remove it and tighten the rest." Iteration is the workflow, not a sign that something went wrong.


A Note on Tool-Specific Prompting

The role-task-context-format framework applies across tools. For guidance tailored to ChatGPT (OpenAI's AI chatbot) specifically - including how to use the system prompt field and multi-turn conversations effectively - see how to write better ChatGPT prompts.

If you use Google Gemini (Google's AI assistant) or another tool, the same core principles apply; the interface differs but the logic does not. Always check each provider's current documentation for any tool-specific features, since capabilities and plan limits change regularly.


Building the Habit: Practice Makes Prompts Better

Reading about prompting helps, but the skill only sticks through practice on real tasks. A useful habit: whenever you get a weak AI response, spend thirty seconds diagnosing why before you rephrase. Was the role missing? Was the context too thin? Did you forget to specify a format?

Over time, this diagnostic reflex becomes automatic - and your first-draft prompts get noticeably stronger as a result.

If you want structured, bite-sized practice, AILE, the Duolingo for AI is built exactly for this: short lessons that teach prompting through hands-on exercises rather than theory, designed for people who want to build the skill without wading through technical documentation.


Frequently Asked Questions

What is a good AI prompt?

A good AI prompt gives the model enough information to produce a useful response on the first or second try. It typically includes a role ("Act as a copywriter"), a clear task ("Write a subject line for this email"), relevant context ("The audience is small-business owners"), and a format instruction ("Keep it under ten words"). The more specific you are, the less the AI has to guess.

How long should an AI prompt be?

There is no ideal length - a prompt should be exactly as long as it needs to be. Short prompts work fine for simple tasks ("Summarise this paragraph"). Complex tasks - drafting a report, writing code, analysing a document - usually need more context, a stated role, and a format instruction. If your output keeps missing the mark, adding detail to the prompt is almost always the right fix.

Can beginners learn how to write AI prompts?

Absolutely. You do not need a technical background. The core skill is clear communication: say who you want the AI to be, what you want it to do, what background it needs, and how you want the answer presented. Most people see a noticeable improvement in their results after practising the role-task-context-format framework on just a handful of real tasks.

Do AI prompts work the same way in ChatGPT and other tools?

The underlying principles - specificity, context, role, format - apply across tools. ChatGPT (OpenAI's AI chatbot), Google Gemini (Google's AI assistant), and similar assistants all respond better to structured, detailed prompts than to vague one-liners. There are minor differences in how each tool handles very long prompts or system instructions, so it is worth experimenting with the specific tool you use most.

What is the most common AI prompting mistake?

The most common mistake is being too vague - writing something like "Help me with my email" instead of "Rewrite this email to a client who missed a deadline. Keep the tone professional but warm, and end with a clear next step." Vague input forces the AI to make assumptions, and those assumptions rarely match what you actually wanted.

How do I improve my AI prompts over time?

Treat every response as feedback. When the output misses the mark, ask yourself what information the AI was missing - then add it to your next prompt. Keeping a short personal library of prompts that worked well is also useful; you can reuse and adapt them rather than starting from scratch each time. Structured practice, such as working through real tasks with a dedicated learning tool, accelerates the process considerably.


Keep going with AILE

Learning AI shouldn't feel like falling behind. AILE, the Duolingo for AI, turns it into short, friendly, hands-on lessons you can actually finish - no jargon, no gatekeeping. Join the waitlist for early access →