Tell It What You Want

It's a Skill, Not a Spell

Two people sat down at lunch on the same Tuesday and typed the same four words into the same kind of AI assistant: help me with my resume.

Call the first one Marcus. (He's made up, and so is everyone else in this book — but you'll recognize him.) Marcus hit enter, leaned back, and waited for magic. What came back was a tidy gray wall of text. A "professional summary" that could have belonged to anyone alive. Bullet points stuffed with words like synergy and results-driven. A skills section that listed "communication" as though that were a revelation. Marcus read about a third of it, felt his shoulders drop, and closed the tab. Overhyped, he thought. Or maybe — and this is the quieter, meaner thought — maybe I'm just not a tech person. He went back to formatting the same Word document he'd been wrestling for a week.

Now call the second one Lena. Same assistant. Same minute, more or less. But Lena typed more. She said she was applying for a job as an operations coordinator at a mid-sized hospital. She pasted in three things she had actually done at her last job — cut the supply-ordering time in half, trained six new hires, untangled a scheduling mess that had been costing weekend overtime. She said she wanted it to sound competent but human, not stuffed with buzzwords. And then she asked for two versions: one straightforward, one a little warmer. What came back wasn't perfect. But it was hers. She could see herself in it. She trimmed a line, swapped a verb, and twenty minutes later she had a resume she was willing to put her name on.

Same tool. Same task. Opposite endings.

The difference wasn't luck, and it wasn't some hidden talent Lena was born with. The difference was that Lena directed the thing, and Marcus just asked it. He treated it like a vending machine — insert request, receive product, complain about the product. She treated it like a sharp new assistant who'd shown up for the first day eager to help and knowing absolutely nothing about her.

That gap is what this whole book is about. And the good news — the genuinely good news — is that it's a learnable gap. Everything Lena did, you can do. None of it required understanding how the machine works. All of it required knowing how to brief it.

Most bad results are a briefing problem

Here's the reframe. Read it twice, because the rest of the book leans on it.

When you get a disappointing answer from an AI assistant, there are three stories you can tell yourself. The first is this technology is overhyped. The second is I'm bad at this. Both are popular, both feel true in the moment, and both are usually wrong. The third story — the useful one — is I didn't brief it well enough.

Think about what actually happened with Marcus. He asked a stranger to help with "my resume" while telling that stranger nothing about the job, nothing about his accomplishments, nothing about the tone he wanted. A brilliant human assistant handed that same assignment would have produced something just as generic — or, more likely, would have stopped and asked him a dozen questions first. The AI didn't ask. It just filled the silence with the most average, most middle-of-the-road answer it could assemble, because average is the safest guess when it has nothing to go on. Bland in, bland out.

This is the heart of it: the intelligence is on tap, but the steering is your job. These tools are astonishingly capable and astonishingly literal. They will take you exactly as seriously as you take the request. Give them a vague wish and they give you a vague draft. Give them a clear brief — what you want, why, who it's for, what "good" looks like — and they become something close to the talented assistant you always wished you had.

You already own this skill, by the way. If you've ever handed work to a new hire, briefed a contractor, told a teenager precisely how you want the kitchen cleaned, or written instructions clear enough that someone could follow them without you in the room — you've done this. Directing an AI is that same muscle. We're just going to point it at a chat box.

This is a driving manual, not an engineering course

You will notice something missing from this book. I am not going to explain how AI works under the hood. Not the math, not the training, not the machinery humming somewhere in a data center.

That's on purpose.

You can be an excellent driver without knowing how an engine converts fuel into motion. You learn the road, the controls, the habits — when to check your mirrors, how to read a wet road, what to do when something feels off. The mechanics matter to the people who build the car. They barely matter to you at the wheel.

This book is the driving manual. It is a collection of moves — things you can copy today, in your next conversation, and reuse for the rest of your life. When a term is unavoidable, I'll translate it the second it appears. Two you'll meet constantly: a prompt is just the request you type — your instructions to the assistant. And context is the background you hand it so it knows what you're talking about — the job, the audience, the example, the constraints. That's most of the vocabulary. Truly. If anyone ever made you feel like you needed a computer-science degree to use these tools well, they were either showing off or selling something.

The Four-A Loop

Almost everything in this book hangs on one simple habit. I call it the Four-A Loop, and it's the rhythm a skilled person runs — usually without naming it — every single time they get good help from an AI. Four steps, in order, and then you go around again.

Aim. Get clear on what you actually want before you type a single word. This is the most-skipped step and the one that quietly ruins the most results. Marcus skipped it entirely; he didn't decide what he wanted, so he couldn't ask for it. Lena did it in her head before her fingers moved: operations coordinator, a hospital, these three wins, competent but warm. Aiming takes thirty seconds and saves you twenty minutes. Chapter 2 is devoted to it.

Ask. Now write the request — specific, briefed, with enough context that a stranger could run with it. This is the part most people think is the whole game, and it's where the side-by-side "weak ask versus good ask" examples in this book will live. There's real craft here, and Chapter 3 is the heart of it. But notice it's only step two. A great ask built on a fuzzy aim is just a well-dressed mistake.

Audit. Check the answer before you trust it. These tools are confident even when they're wrong — they'll hand you an invented fact in the same calm voice they use for a true one. So you look before you leap, and how hard you look depends on what's riding on it. We'll get to a quick way to make that call in a moment, and Chapter 10 turns auditing into a reliable habit.

Adapt. Treat the first reply as a draft, not a verdict. Refine it through conversation — make it shorter, warmer, give me two versions, you missed this — the way Lena nudged hers into shape. And when something works beautifully, save it, so next time you don't start from a blank box. That saving habit becomes your personal toolkit, and it's one of the most valuable things this book will give you.

Aim, Ask, Audit, Adapt. You'll learn each A on its own — Chapters 1 through 4 install the craft — and then, in Chapters 5 through 9, we'll run the full loop on the five jobs people most want help with: writing, summarizing, planning, learning, and thinking things through. Every task chapter is the same loop, applied. Once you feel the rhythm, you'll run it without thinking, the way you signal a turn without deciding to.

The Stakes Test

A quick promise about Auditing, because it's the step that keeps you safe.

Not every answer needs the same scrutiny, and pretending otherwise would just exhaust you. So before you trust anything, ask one question: if this is wrong, what does it cost me?

A brainstormed list of birthday-party themes? If one idea's a dud, you skip it. Glance and move on. A rough first draft you're going to rewrite anyway? Low stakes. But anything you're going to act on — anything touching your health, your money, the law, real numbers, real names, or facts you'll repeat to other people as true — that's high stakes, and it earns real, independent checking. A second source. A qualified human. Your own eyes on the original document.

That single question — if this is wrong, what does it cost me? — is the Stakes Test, and it will travel with you through the entire book. It's the difference between using these tools boldly and using them recklessly.

The ground rules

Four honest things, before we go further. These are permanent. They'll surface again whenever they matter.

First: these tools change constantly. The names, the buttons, the menus, the features — they're different across products and they shift with every update. So I will never tell you "click the blue button on the left," because by the time you read this the button may be green and on the right, or gone. I teach durable habits and judgment — what to do and why — not one product's current layout. When you need the exact steps for your tool, its own official help pages are the place to look. And I will never tell you which assistant is "the best." That's not modesty; it's that the honest answer depends on your task and changes month to month.

Second: everything here is general and well-established. I don't invent statistics, cite studies that don't exist, name experts, or promise that some tool can do a specific trick. Every person in these pages — Marcus, Lena, and all who follow — is openly made up, a stand-in for a situation you might actually face.

Third, and most important: for anything touching your health, your money, or your legal life, an AI is for learning and preparing better questions — never a replacement for a professional. Use it to understand your options, to draft the questions you'll bring to the appointment, to make sense of a confusing letter. Then take the real decision to a qualified doctor, lawyer, or licensed financial professional. The tool doesn't carry the consequences. You do.

Which is the fourth rule, and it's the spine of the other three: you are accountable for what you do with any answer. Not the tool. You. The whole point of this book is to make you the kind of person who can direct these things skillfully and knows exactly when to stop and check — so that the accountability sits comfortably, because you've earned the confidence.

Before you turn the page

Here is your first move, and it's a small one.

Think back to the last time an AI assistant let you down. The bland answer, the wrong answer, the wall of text you closed in disgust. Write down, in one honest sentence, what you actually typed — the real words, not the cleaned-up version.

Then keep it somewhere. A note, a scrap, the back of this book.

We're not going to fix it yet. We're going to do something better. By the time you finish these chapters, you'll be able to look at that sentence and see, plainly, which of the four A's was missing — whether you never Aimed, or Asked too thin, or trusted without an Audit, or quit instead of Adapting. You'll diagnose your own past frustration like a mechanic listening to an engine.

Because that's what this is. Not a spell you either can or can't cast. A skill — ordinary, learnable, yours — and we start learning it now.

Aim First: Know What You Actually Want

Here is a small, uncomfortable truth about getting help from an AI assistant: most of the disappointing answers people show me were ordered, not earned. The person typed a few words into the box, hit enter, and received exactly what they asked for — which was nothing in particular. Then they frowned at the screen and decided the tool was overrated.

I want to catch you before that frown.

In the last chapter we met the Four-A Loop — Aim, Ask, Audit, Adapt — the four moves behind every good result. This chapter is about the first A, and I am going to make a slightly outrageous claim about it. The Aim step happens entirely before you touch the keyboard. It costs about thirty seconds. And it is, by a wide margin, the most skipped step and the most powerful one. Skip it and the other three can't save you. Do it and the others get easy.

So before we talk about what to type, let's talk about what to think.

The blank-minded prompt

Picture handing a task to a brand-new assistant on their first morning. You say, "Can you do something with the Henderson account?" They are eager. They are capable. They have no idea what you mean. So they do the safest, blandest, most defensible thing they can imagine — a tidy summary nobody asked for — and bring it to you with a hopeful look. You're annoyed. They're confused. Neither of you did anything wrong, exactly, except that you never said what "done" looked like.

An AI assistant behaves the same way, only faster and without the hopeful look. When you give it a vague request, it doesn't freeze. It reaches for the average. It has, in a sense, read an enormous amount of ordinary writing on almost every topic, and when you don't tell it which version of "ordinary" you want, it hands you the middle of the pile. The most generic blog post. The blandest email. The listicle that could have been written about anything, for anyone, at any time.

This is the thing to understand and never forget: a blank-minded prompt produces a blank-flavoured answer. If you don't know what a good result looks like, the tool can't either, and it will give you the beige, all-purpose version of your topic. Then — and this is the part I want to spare you — you'll blame the tool for your own unfinished thinking.

Aiming is just thinking done on purpose, before typing. The work of deciding what you want doesn't disappear when you use AI. It only moves. Either you do it deliberately, up front, in half a minute — or you do it accidentally, afterward, by squinting at a wall of generic text and slowly working out what you actually wanted by noticing everything this isn't. The first way is faster. The first way is also the difference between people who say AI saves them time and people who say it wastes theirs.

The three Aim questions

You don't need a worksheet. You need three questions, answered in your head or scribbled on whatever's nearest. I'll give them to you plainly, then we'll watch them work.

One: What's the real outcome I want?

Not the topic. The outcome. "Write about X" is a topic; it's a direction with no destination. "A warm 150-word reply that politely turns down the invitation but leaves the door open" is an outcome. You can hold it in your hand. You'd know it if you saw it.

The trap here is naming the subject and thinking you've named the goal. "Help me with my résumé" is a subject. The outcome hiding inside it might be "rewrite my job history so a hiring manager sees I've managed budgets, in language a hospital would use." Those are very different requests, and only one of them has a target.

Two: Who is it for?

The audience quietly decides almost everything — the words, the tone, the length, what you can assume and what you have to explain. A note to your closest friend and a note to your bank are both "a message about being late on something," and they share not one sentence. The same explanation of a medical appointment reads one way for your worried parent and another for a busy colleague who just needs the date.

When people forget to name the audience, the assistant guesses — and its default guess is a general adult reader on the internet, which is to say, no one in particular. No one in particular is a terrible person to write for. Tell it who's really on the other end.

Three: How will I know it's good?

This is your definition of done, and it's the question that turns a wish into a brief. Length: a tight paragraph, or a full page? Tone: cheerful, formal, sorry, firm? What must it include — a specific date, a name, an apology? What must it avoid — jargon, hype, a hard sell, the word "delighted"? You don't need all of these every time. You need enough that, when the answer arrives, you're checking it against something instead of just reacting to it.

Notice these three questions are not technical. There's nothing about AI in them at all. They're the questions any good editor, manager, or teacher asks before assigning anything to anyone. That's the whole secret of this book in miniature: getting great results is a directing skill, not a technical one. You already have it. You use it every time you ask a competent human for something and get back what you needed. We're just aiming it at the chat box.

The "finished, it looks like…" trick

If three questions feels like a lot to hold, here's a shortcut that does most of their work at once.

Finish this sentence before you type: "When this is done well, it looks like…"

Then describe what you'd be happy to receive — even roughly, even badly. "…a short, friendly email, three sentences, that thanks them and reschedules for next week without grovelling." "…a packing list for a cold-weekend trip with a baby, grouped by bag." "…three sale posts I could paste straight onto the page, no edits."

You are picturing the destination. And a clear picture of the destination is half of a great prompt, because — watch this — the picture is the brief. Read those descriptions again. Every one of them is already most of what you'd type. The instructions wrote themselves the moment you could see the result.

This is why aiming feels like cheating once you get the hang of it. You think you're preparing to write a prompt. You look up and discover you've already written it.

Priya gets a listicle

Let me show you the cost of skipping all this, and the speed of not skipping it.

Priya runs a small bakery on a corner that smells, most mornings, like warm cardamom. She's heard AI can help with marketing. Saturday's slow, so she sits down with her tea, opens the assistant, and types what's honestly in her head:

Help me with marketing for my business.

The reply is instant and enormous. Ten Proven Marketing Strategies for Small Businesses. Define your brand. Leverage social media. Build an email list. Consider influencer partnerships. Each point has a confident little paragraph under it. It is not wrong. That's the maddening part — none of it is wrong. It's also worth nothing to her. It doesn't know she sells bakes, or that her customers are the same forty regulars who've been coming for years, or that what she actually needs, today, is to tell those regulars about Sunday's half-price end-of-day sale. She scrolls. She sighs. She closes the tab. Not for me, she thinks. Overhyped.

The tool did exactly what she asked. She asked for the average, and it gave her the average.

Now rewind. Same tea, same Saturday. This time Priya spends thirty seconds before she types — the three questions, in her head, while the kettle's still ticking as it cools.

Real outcome? Not "marketing." Three short posts for the shop's page, announcing Sunday's sale.

Who's it for? Her regulars. Neighbours. People who already love the cardamom buns and just need a nudge and a time.

How will I know it's good? Friendly, the way she actually talks. Under forty words each so they're easy to read on a phone. No hashtags — she finds them try-hard, and her crowd does too. Each one has to say Sunday and half price after 4 p.m.

She hasn't typed a word to the assistant yet, and look how much she now knows. The vague hope — "help with marketing" — has become a brief she could hand to a person. So that's what she types:

Write three short social posts for my neighbourhood bakery announcing a sale: everything half price this Sunday after 4 p.m. Audience is our regulars, friendly and warm like a chat over the counter. Each post under 40 words, no hashtags, and each one has to mention Sunday and the time.

That request can't help but produce something usable, because there's almost nothing left for the tool to guess. Outcome, audience, and definition of good are all sitting right there in the words. Whatever comes back, Priya can judge it in seconds — too long, soften that line, this one's perfect — because she knows precisely what she's checking against. The hard thinking was already done. By the time she opened her mouth, she was directing, not hoping.

Same person. Same tool. Same thirty seconds she'd otherwise have spent sighing. The only thing that changed was that she aimed.

Aiming also tells you whether to bother

There's one more thing the Aim step does, and it's the part people don't expect: sometimes the honest answer to "what do I want?" is "something an AI shouldn't be giving me."

Aiming isn't only choosing the right words. It's choosing the right tool — and occasionally choosing no tool at all.

A few examples of aims that should send you elsewhere. If what you want is a single hard fact you'll act on — a train time, a dosage, today's exchange rate, whether a shop's open right now — you usually want a source that's current and checkable, not a confident-sounding sentence from an assistant that may be working from old or fuzzy memory. If what you want is a deeply personal decision — should I leave this job, forgive this person, move to this city — no assistant has lived your life, and the thing it's best at, sounding reasonable, is exactly the thing that can talk you smoothly into someone else's answer. And if what you want touches your health, your money, or the law in any way you'll genuinely act on, the aim isn't "get the answer." We'll spend real time on this later, but the short version starts here: for those, an AI is for learning the subject and preparing better questions to bring to a qualified human — never the human's replacement, and never the last word.

The point is that you find this out by aiming, not by typing. Thirty seconds of "what do I actually want, and is this thing the right one to ask?" saves you from the worst outcome of all — not a bad answer, but a confident answer to a question you should have taken somewhere else. We'll sharpen this instinct into a proper tool, the Stakes Test, in its own chapter. For now, just notice that the Aim step has quietly steered you clear of trouble before you ever opened the conversation.

Try this

Before your very next AI request — the real one, the one you'll actually make today or tomorrow — find a sticky note. Or the back of an envelope. Or the notes app. Write three short lines:

• Outcome: what I actually want (the result, not the topic)
• Audience: who it's really for
• Good looks like: length, tone, must-include, must-avoid

Don't aim for elegant. Aim for honest. Then write your prompt — and watch how much of it has already been written for you. The outcome becomes the instruction. The audience becomes the tone. The "good looks like" becomes your checklist for the answer when it lands.

That's the whole move. You point before you shoot. Everything in the rest of this book — the asking, the checking, the back-and-forth that turns a draft into the thing you wanted — gets easier the moment there's a target to hit.

So: aim first. Then we'll talk about how to ask.