How to Write a Good Prompt: Beginner’s Guide 2026

“`html

Don’t forget to visit our AI tools directory!

You open ChatGPT, Claude or Gemini. You type your question. And the answer you get is… disappointing. Too vague, too generic, completely off what you wanted. You close the tab thinking that “AI is overrated”.

The reality? The problem isn’t AI. It’s the prompt.

A prompt is the instruction you give to an artificial intelligence. It’s how you talk to it, how you brief it on its mission. And like with any colleague — human or machine — the quality of the answer depends directly on the quality of the instructions. Give vagueness, you get vagueness. Give precision, you get precision. Simple, basic.

In 2026, knowing how to write a good prompt has become the most important AI skill — the one that separates users who get mediocre results from those who produce professional content in minutes. Professionals who master prompt engineering earn on average 56% more than their peers, according to AI job market data.

This guide gives you the method, concrete examples and mistakes to avoid to go from beginner to effective user, starting today.

What you’ll learn:

• What is a prompt and how do AIs work?
• What structure should you use to write a good prompt?
• What are the most common mistakes and how to avoid them?
• Real examples of bad prompts transformed into good prompts
• Ready-to-copy templates for your common uses

What is a prompt? Understanding to use it better

A prompt is the text instruction you send to an AI model to tell it what it should produce. It can be a simple question, a complex instruction, a role to play or a task to accomplish.

The best way to understand how AI works is to imagine an ultra-competent intern who just arrived at your company. They’ve read everything on the internet, knows thousands of fields, speaks all languages — but knows absolutely nothing about you, your context, your priorities or your preferences. If you tell them “make me a report,” they’ll produce a report — not necessarily yours. But if you give them a precise brief with the context, target, format and expected tone, they deliver exactly what you need.

That’s exactly how a large language model (LLM) like Claude, ChatGPT or Gemini works. These models don’t read your mind. They generate the most probable response based on your instructions. The more precise your instructions, the more relevant the response.

The golden rule: AI delivers results proportional to the quality of instructions received. Nothing more, nothing less.

The beginner’s big mistake: the vague prompt

Before seeing how to write a good prompt, let’s look at what doesn’t work — and why.

Here are the four most common mistakes, illustrated with real examples:

Mistake #1: The request without context

❌ Bad prompt: “Write a cover letter.”

AI doesn’t know for which job, which field, which candidate, which tone. It will produce a generic letter that could fit anyone — so no one in particular.

✅ Good prompt: “Write a cover letter for a project manager position in the renewable energy sector, for a French mid-size company. Highlight team management skills and deadline adherence. Formal and professional tone, 300 words maximum.”

Mistake #2: The fuzzy objective

❌ Bad prompt: “Explain AI to me.”

This prompt can produce a result of 10 lines or 10 pages. For an 8-year-old child or for an engineer. In simple language or technical jargon.

✅ Good prompt: “Explain what artificial intelligence is to a non-technical adult, in 200 words, with two concrete examples from everyday life.”

Mistake #3: Vague adjectives without criteria

❌ Bad prompt: “Write a nice article about digital marketing.”

Nice means nothing to a machine. What’s nice? Humorous? Short? Illustrated with examples? For whom?

✅ Good prompt: “Write a 500-word article on the 3 trends in digital marketing in 2026, aimed at French SME leaders. Accessible tone, without technical jargon. Structure: introduction + 3 sections + conclusion.”

Mistake #4: Asking for multiple things at once

❌ Bad prompt: “Summarize this document, translate it into English and give me a list of key points.”

AI can handle this request, but results will be less precise on each task. Better to separate them into multiple successive exchanges.

✅ Better approach: Start with the summary, validate it, then ask for the translation, then the key points.

The CPTF method: the universal structure of the good prompt

To never have disappointing results again, memorize and apply the CPTF method: Context + Persona + Task + Format.

These four elements form the skeleton of every effective prompt. Not all are always necessary — but the more you include, the more precise the answer.

C — Context

Give AI the basic information it needs to understand your situation. Who are you? Who do you work for? What’s the starting situation?

Examples:

• “I manage an online store of eco-responsible clothing for women aged 25-45.”
• “I’m a master’s student in law and I’m preparing a presentation on GDPR.”
• “Our company just launched new HR management SaaS software.”

P — Persona (the role)

Assign a precise role to the AI. This technique works because language models associate to each role a specific vocabulary, level of expertise and way of reasoning. “You’re an expert in X” activates semantic patterns very different from “Answer my question.”

Examples:

• “You’re a marketing strategy consultant specializing in French micro-enterprises.”
• “Act as a history teacher expert in the French Revolution.”
• “You’re a senior developer with 15 years of Python experience.”

Tests conducted on hundreds of projects show that answer precision increases by 60 to 80% when a role is correctly assigned.

T — Task

Describe the action you’re requesting with a precise verb. Avoid vague wording. Use clear action verbs: write, analyze, summarize, list, compare, explain, translate, correct, rephrase, generate.

Examples:

• “Write a 150-word product description” ✅
• “Do something about my product” ❌
• “Analyze the strengths and weaknesses of this strategy” ✅
• “Talk about my strategy” ❌

F — Format

Specify how you want the answer presented: length, structure, tone, constraints. This is the step beginners most often forget — yet it’s the one that makes the difference between a generic result and one directly usable.

Examples of format clarifications:

• “Maximum 300 words”
• “As 5 bullet points”
• “Professional but accessible tone, without technical jargon”
• “With a catchy title, introduction, 3 H2 sections and conclusion”
• “As a table with 3 columns: advantages, disadvantages, use cases”
• “No emojis, sentences with maximum 15 words”

Complete example: a CPTF prompt in action

Here’s how to assemble the four elements in one complete and effective prompt.

Situation: You manage social media for an artisanal bakery in Lyon and want an Instagram post.

❌ Beginner prompt: “Write an Instagram post for my bakery.”

✅ CPTF prompt:

[Context] I manage social media for an artisanal bakery in Lyon, specializing in sourdough and homemade pastries. Our clientele is local, attached to traditional products and authenticity.

[Persona] You’re a community manager specializing in French food craftsmanship.

[Task] Write an Instagram post to announce our new artisanal king cake, available only in January.

[Format] Maximum 120 words, warm and artisanal tone, maximum 3 emojis, a call-to-action to visit the store. No hashtags.

The result obtained will be targeted, in the right tone, in the right format, directly publishable — where the vague prompt would have produced something generic and unusable.

5 advanced techniques to improve your prompts

Once the CPTF method is mastered, here are five additional techniques that take your prompts to the next level.

Technique 1: Giving examples (the “few-shot”)

The most powerful and most underutilized technique: show AI what you expect rather than describe it. Provide one or two examples of the type of answer you want, and AI will be inspired by the style, structure and tone.

Example:

“Here are my usual LinkedIn post styles: [Post 1: example of tone and structure] [Post 2: example of tone and structure] On this model, write a post about [new topic].”

This technique is particularly effective for reproducing your editorial voice, a brand style or a recurring format.

Technique 2: Asking for step-by-step reasoning

For complex tasks — analysis, problem-solving, decision-making — add this instruction: “Reason step by step before answering.” or “Explain your reasoning.”

AI models tend to want to give a quick answer. By forcing them to reason first, you get deeper analysis and fewer errors, especially on complex subjects.

Example:

“Analyze this business strategy by identifying strengths, weaknesses and risks. Reason step by step for each point before formulating your conclusion.”

Technique 3: Specifying constraints and prohibitions

In addition to saying what you want, say what you don’t want — but phrased positively. Language models process positive instructions better than negations.

Example:

• ❌ “Don’t make something too long”
• ✅ “Keep the answer to 200 words”
• ❌ “No jargon”
• ✅ “Use only vocabulary accessible to a high school student”

Technique 4: Asking for self-criticism

After getting a first response, you can ask AI to evaluate its own work:

“Analyze your previous answer and identify its weak points. Propose an improved version.”

This simple technique significantly improves the quality of deliverables on high-stakes tasks.

Technique 5: Iteration — never stop at the first result

The perfect prompt on the first try doesn’t exist. Iteration is part of the process. Treat your first prompt as a first draft, not as a final version. Progressively refine:

• “That’s good, but make the tone more direct”
• “Shorten the second part”
• “Add a concrete example in the third paragraph”
• “Rephrase the introduction to start with a question”

This dialogic approach — building over multiple exchanges — consistently produces better results than the quest for the “perfect prompt” in a single attempt.

The 7 mistakes to never make

1. Copy-pasting prompts found online without adapting them

Studies show that 78% of collaborators use prompts found on the internet without adaptation, with only 21% successful tasks. A good prompt is always contextualized to your specific situation.

2. Using vague adjectives

Nice, good, pro, cool, interesting, modern — these words don’t guide AI. Replace them with measurable criteria: “3 concrete examples,” “sentences maximum 15 words,” “high school level.”

3. Putting sensitive data in your prompts

Never paste confidential information into a prompt: customer personal data, trade secrets, contracts, passwords. This data transits through AI provider servers and can be used. Always anonymize.

4. Blindly trusting the response

AIs can “hallucinate” — that is, invent information with confidence. Always verify facts, figures and sources provided by AI, especially for technical, medical or legal subjects.

5. Asking for everything in a single prompt

The more overloaded a prompt is, the less precisely each part is handled. Break complex tasks into multiple successive exchanges rather than putting everything in one message.

6. Using negative formulations

Language models process positive instructions better. “Write in 200 words” works better than “Don’t make something too long.” Always formulate what you want, not what you don’t want.

7. Abandoning after the first disappointing result

A bad response isn’t a failure of AI — it’s a signal to refine your prompt. Try again with more context, a more precise role or a better-defined format.

Ready-to-use templates: copy and adapt

Here are six universal templates for the most common uses. Replace elements in brackets with your information.

Template 1 — Blog article:

You’re a web writer expert in [field]. Write a [X]-word article on [topic], for [audience]. Structure: introduction, [N] sections with H2 titles, conclusion. Tone [register]. Include [X] concrete examples. Start with a strong hook.

Template 2 — Professional email:

You’re an expert in professional communication. Write a [type: follow-up / proposal / thank-you] email to [recipient: client / prospect / partner] about [subject]. [Formal/friendly] tone, maximum 150 words. Objective: [desired action]. End with a clear call-to-action.

Template 3 — Social media post:

You’re a community manager specializing in [sector]. Write a [LinkedIn / Instagram / X] post of [X] words for [objective: announce / inform / engage] about [topic]. Audience: [target]. Tone [register]. Maximum [X] emojis. End with [question / CTA / hashtags].

Template 4 — Document summary:

You’re an analytical assistant expert. Summarize the following document in [X] key points, each point in one sentence. Also identify the 3 most important pieces of information and the 2 questions this document leaves unanswered. [Paste your document]

Template 5 — Comparative analysis:

You’re a consultant in [field]. Compare [option A] and [option B] according to these 4 criteria: [criterion 1], [criterion 2], [criterion 3], [criterion 4]. Present the result as a table, then give a recommendation in 3 sentences with the conditions under which each option is preferable.

Template 6 — Idea generation:

You’re a creative expert in [field]. Generate [X] original ideas for [objective] for [target]. For each idea: a catchy title, a 2-sentence description, and why it would be effective. Avoid obvious ideas — prioritize originality and relevance for [specific target].

Prompts according to tools: some nuances to know

Large AI models all respond to the same fundamental principles — context, role, task, format — but have some nuances.

Claude appreciates rich contexts and nuanced instructions. It’s particularly receptive to structured tags (in the style of section headings) and precision of expected tone. On complex subjects, it sometimes questions the request to refine it — which is an asset, not a limitation.

ChatGPT responds well to direct and structured instructions. The phrasing “You’re X, do Y for Z” is very effective. It executes without questioning, which is fast but requires more initial precision.

Gemini benefits from its integration into the Google ecosystem — specify if you want it to perform a real-time web search for recent information.

In all cases, a good prompt created for one of the three tools will work 80-90% on the others.

Conclusion: the prompt, your new superpower

Knowing how to write a good prompt is learning to speak effectively to one of the most powerful technologies of our time. It’s not a skill reserved for engineers or AI experts — it’s a matter of method, clarity and practice.

What to remember:

• An effective prompt rests on 4 elements: Context + Persona + Task + Format (CPTF)
• Precision always beats length: a short, clear prompt beats a confusing wall of text
• Always assign a precise role to AI — this improves quality by 60 to 80%
• Give examples when you want a specific style or format
• Iterate: the first result is a draft, not a final version
• Never copy-paste prompts without adapting them to your context
• Always verify the facts and figures produced by AI

The real skill of 2026 isn’t using AI — everyone does that. It’s knowing how to talk to it better than everyone else. Start with the templates in this guide, experiment, observe what works, and progressively build your personal prompt library. In a few weeks, you’ll transform the way you work.

“`