Writing Better Instructions for AI

You tried an AI, and the result disappointed you: vague response, too long, off-topic. Before concluding that « it doesn’t work, » know one thing: the quality of what an AI produces depends enormously on how you ask it. What is called « prompt » (or instruction) is the message you send to it, and properly formulating it makes all the difference. It’s not magic nor a technique reserved for experts: a few simple principles are enough to transform mediocre responses into useful ones. Here’s how to write better instructions to an AI to get more out of it without jargon or complicated recipes.

Why the way you ask makes a difference

A generative AI (our guides to understand generative AI) does not guess what you have in mind: it responds to what you WRITE, based on what you give it. If your request is vague, the response will be as well; if it is precise and well-framed, the response gains in relevance. It’s exactly like assigning a task to someone: the clearer your instructions, the better the result. The ingredients of a good instruction. THE CONTEXT: give the AI the information it needs (what it’s about, for whom, in what context); it doesn’t know your situation, so you have to describe it. THE SPECIFIC TASK: clearly state what you want it to do (explain, summarize, write, correct, compare, list); a precise action verb is better than a vague request. THE DESIRED FORMAT: specify the expected form (a list, a paragraph, a table, a few lines, a structured text); otherwise, the AI will choose for you, often too long. THE TONE AND AUDIENCE: state the register (formal, simple, friendly) and who the result is for (a child, a client, an expert); the AI adapts. THE LIMITATIONS: length, what to include or avoid, the language; these limits guide the response. USEFUL EXAMPLES if applicable: showing an example of what you want helps the AI aim accurately. What’s the point? To get RELEVANT responses the first time; to SAVE time (fewer back-and-forths); to fully exploit the POTENTIAL of the tool; and to avoid the frustration of « it doesn’t work. » The main idea: an AI is as good as the request you make of it. Most disappointments come from too vague instructions, not from the tool itself. By learning to formulate what you want—context, task, format, tone, limitations—you radically transform the quality of the results. It’s not an expert skill: it’s common sense applied, the same we use to clearly assign a task to someone. And it improves quickly with practice.

Principles of a good instruction

Here are the concrete principles to better interact with an AI. BE PRECISE AND CONCRETE — instead of « talk to me about sports, » say « explain in three simple points the benefits of walking for a sedentary 60-year-old »; precision guides the response. GIVE CONTEXT — explain the situation: « I’m preparing an email to a dissatisfied client, » « I’m a total beginner, » « it’s for a school presentation »; the AI adapts its response to your case. STATE THE DESIRED FORMAT — « answer in a bulleted list, » « in a short paragraph, » « in five sentences maximum, » « in the form of a table »; you avoid long blocks of text. PRECISE THE TONE AND AUDIENCE — « in a simple tone, for someone who knows nothing about it, » « professionally, » « as if you were explaining to a child. » DECOMPOSE COMPLEX TASKS — for an important task, proceed step by step rather than asking everything at once; start broad, then refine. ITERATE: THE DIALOGUE IS THE KEY — the first response is rarely perfect; that’s normal; react: « shorter, » « simpler, » « develop the second point, » « give an example, » « this is not what I wanted, rather this »; it’s by refining that you get the right result. USE AN EXAMPLE IF YOU HAVE ONE — « here’s the style I want, » « follow this example »; the AI mimics a model very well. ASK WHAT YOU DON’T KNOW — if you don’t know how to formulate, ask it: « what information do you need to help me? »; it can guide you. RESCUE AND REFRAME — if the response strays, get it back on track: remind what you want, what to avoid. A few TIPS: don’t be afraid to SPEAK NATURALLY (no need for magic words, write as if to a person); CRITICALLY REREAD the result (the AI can make mistakes); and PRACTICE (progress quickly by observing what works). In summary: be precise and concrete, give context, state the format, tone, and audience, decompose complex tasks, and especially ITERATE — the dialogue and refinement are the real secret. Formulating a good instruction is simply explaining what you want well; and that’s learned by practicing.

The right reflex. Remember a simple image: talking to an AI is like assigning a task to a competent person who knows nothing about your situation. You wouldn’t ask an assistant « write me a text » without further details; you would give the context, the goal, the desired form. Do the same. The most profitable recipe lies in four elements to slip into your request: THE CONTEXT (what it’s about, for whom), the SPECIFIC TASK (a clear verb: explain, summarize, write, compare), the DESIRED FORMAT (a list, a short paragraph, five sentences, a table — otherwise, you’ll get a block of text), and the TONE or AUDIENCE (simple, professional, « as if to a child »). An example is worth a thousand explanations: compare « talk to me about nutrition » (vague, generic response) and « give me five simple tips for better eating every day, in a list, for someone who doesn’t have time to cook » (precise, useful response). But the real secret, that few people exploit, is ITERATION: the first response is just a starting point; dialogue! « shorter, » « simpler, » « develop this point, » « not like that, rather like this. » It’s by refining, exchange after exchange, that you get exactly what you want. And above all, write NATURALLY: no need for magic formulas or jargon; explain as if to a person. Finally, always critically re-read the result. Asking well is not an expert skill: it’s common sense, and it improves very quickly with a bit of practice.

Common mistakes and limitations

Some traps often come up when starting with AI. A VAGUE REQUEST: this is the number one mistake; « talk to me about X » gives a generic and disappointing response; be more precise about what you really want. EXPECTING PERFECTION ON THE FIRST TRY: many give up after an imperfect response, when iteration through dialogue is actually the key; iterate. WANTING EVERYTHING AT ONCE: an overloaded request swamps the AI; decompose. THINKING YOU NEED MAGIC WORDS: no, write naturally; clarity counts, not secret formulas. FORGETTING THAT THE AI KNOWS NOTHING ABOUT YOU: it has no access to your context, documents, situation, unless you give it. But beyond the formulation, keep in mind the LIMITATIONS of the AI itself. It can make MISTAKES: a good instruction improves relevance and form, but does not guarantee accuracy; the AI can invent facts, make calculation errors, assert an error with confidence (our guides on the limitations of AI). It has no JUDGMENT: it produces a plausible text, not a truth; keep your critical mind and the final word. CONFIDENTIALITY: better formulation sometimes pushes you to give a lot of context; be careful not to deliver very personal or sensitive information without thinking (our guides on AI and privacy). DEPENDENCE: using AI well is useful, but keep your own capacities of reflection and writing; the AI assists, it does not think for you. Finally, each TOOL has its specifics: the principles are universal, but some AI react a bit differently; observe and adapt. In summary: avoid vagueness, don’t expect immediate perfection (iterate), don’t overload, write naturally, and give the necessary context; but remember that a good instruction makes the AI more relevant, not infallible: always verify, keep your critical mind, and protect your data. Proper formulation multiplies the tool’s effectiveness; it does not replace your vigilance.

Be careful: proper formulation makes the AI more relevant, not infallible. It would be tempting to believe that with the « right » instruction, the AI will always give a correct response. That’s false, and it’s a trap. Improving the way you ask improves the RELEVANCE and form of the response (it better targets your need), but it does not guarantee its ACCURACY. An AI can, even with a perfect instruction, invent facts, make calculation errors, cite a non-existent source, or assert a falsehood with the utmost confidence. Do not confuse a well-formulated and well-presented response with a true one: always verify what is factual or important, the AI has no judgment, it produces the plausible, not the certain. A second vigilance: to avoid giving a « good context, » one is tempted to reveal everything; be careful not to deliver very personal, professional, or sensitive information to an online tool — give the useful context, not your intimacy or confidential data. Also beware of DEPENDENCE: learning to dialogue well with the AI is a useful skill, but keep your own capacities of reflection, writing, and analysis; the AI should assist you, not think for you. Finally, do not seek « magic formulas »: sellers of miracle recipes to « master AI » ride on a myth; the true principles are simple, free, and common sense (context, task, format, iteration). Proper formulation will greatly benefit you; but always, between you and the AI’s response, keep your critical mind — it’s him, not the perfect instruction, who protects you from its errors.

Frequently asked questions

Do you need to learn a special language to talk to an AI?

No, absolutely not. One of the most widespread misunderstandings is that you need to know « magic formulas, » a jargon, secret techniques. In reality, modern conversational AIs are designed to understand NATURAL LANGUAGE: you talk to them as you would to a person, in everyday French. What matters is not mysterious keywords, but the CLARITY of your request: say precisely what you want, give the context, specify the desired format. Exactly as when you explain a task to someone. Beware of the countless « magic prompt guides » sold here and there: they ride on a myth. The true principles boil down to a few common-sense ideas (be precise, give the context, state the format, iterate), free and easy to apply. The best way to progress is not to learn a list of formulas, but to PRACTICE: try, observe what gives good results, adjust. With a few uses, you will naturally develop the right reflexes. Talking to an AI is a dialogue in ordinary language: the only « skill » required is to be able to clearly explain what you need.

Why does the AI give me off-topic responses?

Most often, because the request was too VAGUE or lacked CONTEXT. The AI does not guess what you have in mind: it answers what you write, literally. If you ask « give me ideas, » it does not know what, for whom, or with what purpose: the response will be generic. Remedies: PRECISE your need (the exact subject, the objective, the constraints); give the CONTEXT (your situation, for whom, what use); indicate the FORMAT and desired length. And above all, if the first response is off-topic, do not give up: REFRAME. Say what did not work and what you really wanted: « no, I wanted ideas for a child’s birthday, not for a wedding, » « more concrete, » « in a different style. » This adjustment dialogue is normal and effective: it is how we converge towards the right result. It also happens that the AI makes a mistake or invents something: in this case, a better instruction helps, but keep in mind that it is not infallible and verify what is important. In summary, an off-topic response is almost always a sign that the request needs to be clarified or reframed: reformulate with more context and precision, and the quality improves immediately.

How to improve a response that doesn’t suit me?

By dialoguing, because that’s the secret: the first response is never the final one. Rather than starting from scratch or giving up, REACT to what the AI produced, with adjustment instructions. Some working examples: « it’s too long, summarize in five lines »; « too complicated, explain more simply »; « develop the third point more »; « give me a concrete example »; « keep the idea but change the tone, more warm »; « this is not what I was looking for, I wanted this instead. » The AI keeps the thread of the conversation and refines its response based on your feedback. You can chain as many adjustments as necessary until you get what you want. It’s exactly like working with a collaborator: you clarify, correct, and refine together. Don’t hesitate to ask for SEVERAL versions (« propose me three different formulations ») to choose from, or ask why it chose a particular option. This iterative approach is much more effective than trying to find the perfect request the first time. By getting into the habit of dialoguing and adjusting, you will almost always, in a few exchanges, get a result that really corresponds to what you wanted.

What to remember

Most disappointments with an AI come not from the tool itself, but from how it is asked: an AI is as good as the instruction it receives. The good news: better formulating does not require jargon or expert techniques, only common sense — the same we use to clearly assign a task to someone. The ingredients of a good instruction: THE CONTEXT (what it’s about, for whom — the AI knows nothing about your situation), the SPECIFIC TASK (a clear verb: explain, summarize, write, compare), the DESIRED FORMAT (a list, a short paragraph, five sentences, otherwise beware of blocks of text), and the TONE or AUDIENCE. Compare « talk to me about X » (vague, disappointing) and a precise and well-framed request (useful) to see the difference. But the real secret, too little exploited, is ITERATION: the first response is just a starting point; dialogue, adjust (« shorter, » « simpler, » « develop this point, » « not like that, rather like this »); it’s by refining that you get exactly what you want. Write NATURALLY (no magic formulas: beware of « miracle guides » that ride on a myth), decompose complex tasks, and give an example if you have one. However, keep in mind a key limit: better formulating makes the AI more RELEVANT, never INFAILLIBLE; even with the perfect instruction, it can invent, make mistakes, assert an error with confidence; always verify what is factual or important, and keep your critical mind — it’s him, not the instruction, who protects you from its errors. Also beware of delivering too personal data under the pretext of a « good context. » By learning these few principles and practicing, you will radically transform what the AI brings you: from mediocre responses to a real assistant. The difference is not in the machine; it is in the clarity of what you ask of it.