How to Use AI Properly: The Beginner's Guide That Actually Works (2026)

Most people use AI the way they use a search box: type a few words, take whatever comes back, feel vaguely disappointed. Then they conclude "AI is overrated." It isn't — they're using a power tool like a hand tool.

Using AI properly is a skill, and it's a small one. Five habits cover almost everything. This guide teaches them with examples you can paste into ChatGPT, Claude, Gemini or our free brain right now. No jargon, no theory you don't need.

The one idea behind all of it

An AI model answers the question you actually asked, not the one in your head. When you type "write a cover letter", the model has to guess the job, your experience, the tone, the length and what the reader cares about — and it guesses generically, because generic is the safest guess.

Everything below is a way of replacing guesses with facts. Every fact you give it removes a guess. Every guess you remove makes the answer more yours.

Habit 1 — Brief it like an expert, not like a search engine

Imagine hiring a brilliant freelancer for one hour. You wouldn't say "cover letter". You'd say who you are, what job, what you've done, what tone you want, how long it should be. The AI needs exactly that brief.

A useful brief answers five questions:

1. Who should the AI be? — a role sets the vocabulary and standards. "You're a recruiter who has read 5,000 cover letters."
2. What's the situation? — the context the expert would ask for first. "I'm a nurse with 6 years in paediatrics applying to a children's hospice."
3. What exactly do you want? — one clear deliverable. "A three-paragraph cover letter."
4. Any rules or limits? — "Under 250 words. No clichés like 'passionate'. Mention my palliative-care certificate."
5. How should it look? — "Plain text, ready to paste."

That's the whole structure — we call it RCTCO (Role, Context, Task, Constraints, Output). You don't need all five every time. You need the ones that would otherwise be guessed.

Before: "Write a cover letter for a nursing job."

After: "You're a recruiter who has read 5,000 cover letters. I'm a nurse with 6 years in paediatrics applying to a children's hospice (job ad below). Write a three-paragraph cover letter under 250 words that leads with why I want hospice work specifically, mentions my palliative-care certificate, and avoids clichés like 'passionate' and 'dynamic'. Plain text. Job ad: [paste]."

Same tool, same model — completely different result.

Habit 2 — Give it the real material

The biggest mistake after vagueness is paraphrasing. People describe their document instead of pasting it, describe the error instead of copying it, summarise the email instead of including it. The model can only work on what's in front of it.

• Fixing writing? Paste the actual draft.
• Debugging? Paste the exact error and the code.
• Replying to someone? Paste their message.
• Analysing numbers? Paste the table (with headers).

Two things to strip first: names and private details you wouldn't want in a stranger's notebook, and anything you're not allowed to share. Round numbers and placeholders work fine for the rest.

Habit 3 — Ask for the shape of the answer

The default output is a wall of prose with bold headings. You can ask for anything else, and it changes how useful the result is:

• "Give me a table with columns: option, cost, risk, my next step."
• "Numbered steps, one sentence each, no explanations."
• "Three versions: safe, bold, and weird."
• "Under 120 words."
• "End with the single most important thing to do first."

Format instructions are the cheapest upgrade in all of prompting. Five words — "as a table, under 200 words" — routinely turn a skim-and-forget answer into something you act on.

Habit 4 — Treat it as a conversation, not a vending machine

The first answer is a draft. The real value is in the follow-up:

• "Shorter." / "More specific." / "Less formal."
• "You assumed X — actually Y. Redo it."
• "Now do the same for the second product."
• "What would a sceptical expert say is wrong with this?"
• "Give me the three biggest risks you glossed over."

The last two are the most under-used prompts in existence. Asking the model to critique its own answer against a standard — accuracy, completeness, tone — and then improve it, reliably beats a single perfect prompt. We wrote a whole guide to it: self-critique prompting.

If your tool forgets the earlier turns, that's a tool problem, not a you problem. (Ours remembers the whole conversation, including the follow-ups.)

Habit 5 — Never trust, always verify — but verify the right things

AI models are confident and sometimes wrong. The trick is knowing what kind of wrong to expect:

• Facts, numbers, dates, citations — verify every one you'll rely on. Models invent references that look completely real. Ask "which of these are you unsure about?" and it will usually tell you.
• Reasoning and structure — usually strong; you'll spot gaps by reading critically.
• Judgment about your situation — it doesn't know your boss, your market or your kid. Use it to think, then decide yourself.

For anything medical, legal, financial or safety-related: use AI to prepare questions and understand terms, then talk to a professional.

Putting it together: a five-minute routine

1. Say who the AI should be and what the situation is (two sentences).
2. Say exactly what you want and how it should look (one sentence each).
3. Paste the real material.
4. Read the answer, then ask for the change that would make it yours — shorter, sharper, corrected.
5. Ask it to critique its own answer once. Verify any facts you'll use.

That routine is what "using AI properly" means. It's not a secret prompt; it's a habit of being specific and staying in the loop.

Where to go from here

• Get prompts for your job. We organised 3,600+ prompts by field — students, marketing, developers, investing, job search, teachers and 37 more. Every prompt is full text, free.
• Let the structure write itself. The Studio asks you the five questions from Habit 1, shows you the real prompt as you type, and runs it — on your phone, one call.
• Learn it properly. 36 short courses drill one habit at a time with exercises and quizzes; five are free.

FAQ

What is the best way to use AI as a beginner?

Start with one real task you do every week — an email, a plan, a summary — and brief the AI the way you'd brief a person: who it should be, the situation, exactly what you want, any rules, and the format. Then improve the answer with follow-ups. You'll be better than most users in a day.

Which AI should I use: ChatGPT, Claude or Gemini?

For everyday work they're closer than the marketing suggests, and the habits in this guide work on all of them. Pick the one your workplace or school allows and learn to brief it well — that matters far more than the brand.

How do I get AI to stop giving generic answers?

Give it what it's guessing: the audience, the situation, your actual material, the format and the rules. Ban the clichés explicitly ("no 'in today's fast-paced world'"). Then ask it to critique and tighten its own draft.

Is it safe to put my information into AI?

Strip names, IDs and anything confidential, check your organisation's policy, and prefer tools that don't train on your data. For most personal tasks — planning, writing, learning — round numbers and placeholders are all the model needs.

Can AI replace learning the skill?

It can replace the typing, not the judgment. The people getting the most from AI are the ones who know what a good answer looks like in their field — which is exactly why it pays to learn to brief it properly.