How to Write Prompts So AI Gets It Right the First Time: A 5-Part Formula

By the end of this article you'll have a clear scheme for how to frame a request for any task, a set of ready-to-copy prompts, and an understanding of why your answers used to come out empty. No technical background needed – I explain it so any expert who can type a message in a chat app will follow.

Section 01It's almost always your question

The most common beginner complaint sounds like this: "I tried your AI, it spits out some kind of fluff, nothing useful." Almost every time I ask to see what the person actually wrote, the picture is the same. In the window there's one short line: "write a post about my course" or "make me a content plan." That's it. The AI got a generic question and honestly gave a generic answer.

Imagine you hired a sharp assistant, sat them down in front of you, and said: "make it good for me." What would they do? At best, ask you to clarify; at worst, invent what you even meant and miss the mark. An AI rarely asks back and more often fills in the blanks itself. And it fills them in with the average, the most common and safe option. That's where the bland fluff comes from: you handed it the right to decide for you, and it decided the gray way.

The point is that an AI doesn't read minds and doesn't know your context. It doesn't know your niche, your audience, or your platform: that you're a psychologist writing for moms on maternity leave, that the post needs to be short and for Telegram, not a longread for a website. All of that lives in your head. Until you say it out loud, for the AI it doesn't exist. It answers literally what's written, and everything still in your head simply isn't there for it.

From that comes a simple rule that flips the whole way you work with an AI. A bad answer is almost always a signal that the prompt was too generic. And that's great news: the prompt is entirely in your hands. You can't rebuild the AI itself, but you can rewrite your question to it in a minute. There's even a fast test for vagueness: if you can hand your request, word for word, to ten different specialists and they'd each read it a different way, it's too fuzzy, and the result will be just as fuzzy.

The rest of this article is about one thing: how to stop asking the AI generic questions and start asking specific ones. For that there's a simple formula, and that's where we head next. But first, let's agree on the words.

Section 02What a prompt is and why one line isn't enough

A prompt is your request to an AI. Any text you send it to get a result. The word comes from "prompt" as in a cue, but really it's just the assignment you set. You typed "translate this email into Spanish" – that's a prompt. You dictated three paragraphs of detail by voice – also a prompt, just a fuller one.

The key idea to accept from the start: an AI answers to the exact amount you put into the request. Put in one line, get an average answer to one line. Put in a paragraph of detail, get an answer fitted to your details. Between those two answers there's a chasm, and it costs you only a couple of extra sentences in the request.

Why that happens is clear from an everyday example. You message a handyman you know: "make me a table." He'll make some table, to his own taste, out of whatever's around. Now: "make me a writing desk for a laptop, light wood, four feet wide, with one drawer on the left, Scandinavian style." The second order the handyman builds the way you need it, because you made all the decisions for him. With an AI it's exactly the same. Every detail in the prompt is one decision you made yourself instead of leaving it to the machine to guess.

Notice something important: a long prompt doesn't mean a good one, and a good one doesn't mean a clever-sounding one. You don't need to write the AI an essay or sprinkle in fancy words. You need to spell out a few specific things it can't know on its own. Which ones, the five-part formula shows. Treat it as a checklist of meanings rather than a rigid form to fill in line by line: run your eyes down it and add to the request whatever's missing.

Section 03The five-part formula

Here are the five bricks a strong prompt is built from. Run through them like a checklist: role, task, context, format, examples and constraints. Let's break down each one with examples so it's clear what to actually write.

You don't have to fit all five parts into every request. For a small thing like "reword this sentence more simply," just the task is enough. But the more important and complex the result, the more bricks are worth laying. General rule: if the answer came out wrong, look at which part was missing. Bland and generic – add context. Wrong tone – add format. Missed the meaning – sharpen the task. The formula works both as a way to write a prompt and as a way to fix a broken one.

Section 04Before and after: three real examples

Theory comes alive in specifics. Let's take three typical expert tasks and see how the same request changes once you apply the formula. On the left – how a beginner usually writes it. On the right – the same intent, broken into the five parts.

Example 1. A post for a Telegram channel

What you get from a request like that: a faceless text with generic lines like "AI is changing the world" and "don't miss out," without a single detail about your course or your people. Impossible to read, embarrassing to publish. Now let's break it into the formula.

The difference is obvious. The second prompt has a role (a copywriter with no clichés), a task (a post with a call to a consultation), context (who the author is, who the audience is, what they fear), format (length, tone, structure, platform), and constraints (no markdown, no invented numbers). The AI now writes your exact post for your exact people. Generic text about nothing stops coming out of it. Edits after a request like that are minimal.

Example 2. Mining client reviews

To a request like that the AI hands you a vague "the reviews are mostly positive, clients are happy." Thanks, you already knew that. Let's add the formula and a concrete goal.

Now the AI mines real material out of the reviews: your audience's language, the recurring results, ready quotes. What comes out are finished pieces for the landing page that you just drop in. Context (who you are, what the landing page is for) and format (three clear blocks, verbatim quotes) did the work.

Example 3. Ideas for reels

You'll get a dozen generic themes along the lines of "show your behind the scenes" and "do a myth breakdown," which suit anyone and hook no one. Let's sharpen it.

The difference is that now the ideas grow out of a specific audience's life, not out of thin air. Hook, gist, call to action – a ready draft for each video. The constraint "no generic words about health and motivation" pulled its own weight: it cuts exactly the bland layer the AI fills the void with when it's short on context. By the way, if you want to assemble a whole month of scripts in one pass, I laid that out in a separate article on prompts for experts.

Section 05Ready prompts for an expert's tasks

So you don't have to assemble the formula from scratch every time, keep a few templates for common tasks on hand. Below is a short map: which task, and what you must spell out in the prompt so the answer is usable. After that – three ready templates to copy, where you just drop in your own details.

And here are three universal templates. Copy them, drop your own details into the square brackets, and send it to the AI.

Template for a post.

You're a copywriter who writes living texts with no cliches.
Write a post for [platform], the goal is [what the reader should do].
About me: [who you are, your topic].
Audience: [who they are, what they care about, what they fear].
Format: [length], tone [which], address [casual/formal],
structure: [for example: hook, story, idea, call to action].
Constraints: no corporate speak, no long dashes, [your no-go list].

Template for breaking down material (reviews, chats, a call transcript).

You're an analyst. Here's my material: [paste the text].
Break it down and pull these for me: [what exactly you need, point by point].
Context: why I need this - [goal], I'm [who you are].
Output format: [how many blocks, with subheads, quotes verbatim].
Don't make anything up, work only with what's in the text.

Template for ideas and variants.

You're a [role for the task: scriptwriter / marketer / editor].
Come up with [how many] variants of [what: headlines / reel ideas / post topics].
About me and the audience: [niche, who the reader is, their situation].
The criterion for a strong variant: [what "strong" means to you].
For each variant give: [what to show for each one].
No generic words or cliches, only specifics for my audience.

Once you build a library like this for your own tasks, you stop writing prompts from scratch and just pull the right template. That saves hours a month. If you want to figure out which tasks are worth handing an AI first, check out the overview of what you can do with AI without a single line of code.

Section 06How to refine an answer in the chat

Even a strong prompt doesn't always land the ideal on the first try. That's normal. The beginner's mistake is, on seeing the wrong thing, to wipe it all and write the prompt again. That way you lose the ground you already gained. The right move is to refine it in the same chat, because the AI remembers the earlier messages and holds the context of the conversation.

It works like a conversation with a live assistant. The answer came out long – you write "make it about a third shorter." Too dry – "add a living example from my audience's life." Wrong angle – "the first paragraph is weak, start with a question that hits the pain." The AI doesn't take offense at edits and redoes exactly the piece you asked about, keeping the rest. Two or three tweaks like that usually get the text where you need it.

A few phrasings worth keeping in mind as working editing tools:

• "Give me three options for the first paragraph to choose from" – when you don't like the opening but can't be bothered to explain what you want.
• "Cut the corporate speak and cliches, keep the meaning" – when the text sounds stiff and official.
• "Better, but you lost the warmth from the last version, bring it back" – the AI remembers the previous version and merges the best of both.
• "Explain why you picked that headline" – when you want to understand the logic and refine it deliberately.
• "Ask me 5 questions that would help you write this better" – a powerful move: the AI pulls the missing context out of you itself.

That last move deserves its own word. If the task is hard and you don't fully know what you want yourself, don't agonize over the prompt – flip the order. Describe the task as it is and ask: "before you write, ask me questions whose answers will help you do this precisely." The AI asks about the audience, the goal, the tone – and you gather the context in a dialogue instead of in one long message. It's also a way to learn to write prompts: the AI's questions show you which bricks you usually leave out.

Section 07Where to write prompts: Claude or ChatGPT

The five-part formula works the same in any AI – it's a universal skill that carries over to any service. Still, people often ask where exactly to write. Short version: both main models, Claude and ChatGPT, handle detailed prompts well, so you can start with whichever is already at hand.

For working with text I find Claude more convenient. It holds a long context carefully, doesn't lose details from the start of the conversation, follows a format well, and writes in a livelier voice with less corporate stiffness. For an expert who writes posts, articles, and emails every day, that's a noticeable difference. ChatGPT is stronger on some other tasks and is simply more familiar to many. There's no right answer here: try both on your real tasks and stay with the one you enjoy more.

One practical note on access. In some countries Claude and ChatGPT don't open directly: you'll need a VPN set to the US or EU, otherwise the login won't go through. If you haven't set up Claude yet, registration starts at claude.ai – turn on the VPN if you need it, sign up with your email, and from there you talk to it like a normal conversation partner.

The important thing is not to get stuck choosing a tool. People will spend weeks deciding which AI is "better" and never actually start. The gap between the services is smaller than the gap between a weak and a strong prompt inside the same service. Pick any one, master the formula, and you're already ahead of most.

Section 08Five mistakes that break a prompt

To avoid the usual rakes, run through this list. Each mistake is a missing part of the formula or an attempt to skimp on clarity.

The common root of all five mistakes is one: trying to get a lot while putting in a little. The AI honestly reflects what you put into it. Put in clarity and context – you get value. Put in one lazy line – you get a lazy answer. So the best strategy is simple: treat the prompt like a brief for a good doer you respect, and it'll answer you in kind.

Section 09Where to start today

So the formula doesn't stay theory, here are three steps you can genuinely do in the next hour with no prep at all.

First – rewrite one of your old requests. Think of a task where the AI recently handed you fluff. Open it again and ask the same question, but by the formula: role, task, context, format, constraints. Compare the two answers side by side. That difference is more convincing than any article and shows on the spot that the problem was the prompt.

Second – build your three templates. Take the three tasks you do most often: say, a post, a review breakdown, ideas. Assemble a prompt for each by the formula and save it to your notes. From that moment you're not writing from scratch anymore – you just drop details into the ready thing. In a week you'll have a personal library of working requests.

Third – practice refining. On any AI answer, make at least one deliberate edit in the chat: "shorter," "add an example," "cut the corporate speak." Get used to the idea that the answer is a draft you can and should talk to. This skill pays off every single day.

And keep the main thought of the whole article in mind. The AI doesn't decide for you whether the result comes out good. You do – by how you set the task. A bad answer almost always means the prompt was too generic, and that gets fixed in a minute. Once you've mastered the simple five-part formula, you stop depending on luck and start reliably getting what you need out of the AI. From there it turns into a habit: you barely think about the bricks anymore and just set clear tasks, getting strong answers. That's where real work with AI begins – with the ability to explain precisely what you want. No magic button needed for that.

If you're just walking into the topic and want a general route for where an expert should even begin, take a look at the article on getting found in AI search. And once prompts become second nature, the logical next step is to fold them into ready Claude skills for your texts, so you don't have to type a long request every time.

RecapThe whole thing in a minute

• Bad answer = bad prompt. To a generic question the AI gives a generic answer. The question is entirely in your hands, and it gets fixed in a minute
• A prompt is a brief. The AI answers to the exact amount you put in. More detail – closer to what's in your head
• The five-part formula: role → task → context → format → examples and constraints. A checklist of meanings, not a rigid form
• Context matters most: who you are, who the audience is, what the goal is. One paragraph about your people makes the answer yours, not generic
• Refine in the chat: don't rewrite from scratch – edit piece by piece – "shorter," "add an example," "cut the corporate speak"
• You can ask the AI to write the prompt for you or to ask you questions before it starts
• The formula works in both Claude and ChatGPT. If they don't open in your country, use a VPN set to the US or EU