← Back to Blog

How to actually start using AI

Start here · No jargon

How to actually start using AIA guide for people who feel behind

No coding. No subscription. Three tools and about twenty minutes — plus an honest list of the things it gets confidently, completely wrong.

My aunt asked me about this at a family dinner last month. She'd been putting the question off for a year and finally admitted it, a bit embarrassed, somewhere between the main course and dessert.

"Everyone talks about it like I should already know. I don't want to ask a stupid question."

She'd been using a computer since before I was born. She was not behind on technology. She was behind on vocabulary — on a wall of jargon built by people who benefit from it sounding complicated.

So this is the answer I gave her, written down. No prompt engineering. No transformers. Nothing you need to memorise.

You are not behind

The honest truth is that most people who talk loudly about AI are using it for about three things: writing emails, summarising documents, and looking things up. That's it. That's the secret club.

There is no accumulated knowledge you missed. Someone who started in 2023 has almost no advantage over someone starting today, because the tools have changed so much that their head start mostly expired.

If you can write a text message, you already have the only skill required. Everything else is knowing what to ask for.

What AI actually is, in one paragraph

Modern AI chat tools are, at their core, extremely good pattern-matchers trained on enormous amounts of text. You give them words, they predict what words should come next, and they do it well enough that the result reads like thinking.

That sounds reductive, and people argue about whether it's the whole story. But hold onto it, because it explains both why AI is so useful and why it fails in the specific, strange ways it does. It produces what a good answer looks like. Usually that's the same as a correct answer. Not always.

The three tools worth starting with

Ignore the lists of fifty. You need one chat assistant, and you should use its free version.

ChatGPTFree tier

The default, and by an enormous margin the most used. Widest range — text, images, voice conversations. If you only ever try one thing, try this.

ClaudeFree tier

Better at long documents and writing that doesn't sound robotic. If you want to paste in a twelve-page PDF and ask questions about it, this is the one.

Google GeminiFree tier

Best if your life already lives in Gmail and Google Docs, because it works inside them rather than beside them.

All three are free to start and none require a card. Pick one and ignore the others for a month. Switching between them before you know what you want is how people end up with three half-learned tools and no habit.

How to talk to it

People call this "prompting" and dress it up as a technical discipline. It isn't. It's the same skill as briefing a new colleague who is clever, fast, and knows nothing whatsoever about your situation.

Watch the difference.

Vague — you'll get something generic Write an email to my landlord.
Specific — you'll get something usable Write a short, polite but firm email to my landlord. The heating has been broken for 11 days. I've called twice and nobody came. I want a repair date in writing this week. Keep it under 120 words. Don't be aggressive — I want to stay on good terms.

Same tool. Completely different result. The second one gave it the facts, the tone, the length and the goal.

The four things worth including

What you want — an email, a summary, a plan, a list
Who it's for — your boss, a customer, a seven-year-old
The facts it can't know — dates, names, what already happened
How long — "under 100 words" changes the output more than anything else

And the one almost nobody tries: tell it what was wrong and ask again. "Too formal, make it warmer." "Cut it in half." "You made up the part about the deposit — remove it." It's a conversation, not a vending machine, and the second answer is usually far better than the first.

What it gets wrong — read this part twice

This is the section most beginner guides skip, and it's the one that matters most.

The single most important thing to know

AI makes things up, and it does it confidently. It will invent statistics, cite books that don't exist, quote people who never said it, and give you a legal deadline that's wrong — all in the same self-assured tone it uses when it's right.

There is no change in voice when it starts guessing. You cannot tell from how it sounds.

This isn't a bug someone will patch next year. It comes from what the thing fundamentally is: a system producing text that looks like a good answer. Most of the time, looking right and being right coincide. Sometimes they don't.

The practical rule: anything with a number, a name, a date, a law, or a medical fact needs checking somewhere else. Anything about how to phrase something, how to structure something, or how to get unstuck — trust it much more freely. It's excellent at shape and unreliable on fact.

Four other honest limitations

It doesn't know what happened last week unless it searches. Ask about recent news and you may get a confident answer from a year ago.
It doesn't remember you between conversations unless memory is switched on. Each new chat starts from nothing.
It agrees with you too much. Push back and it will often fold immediately, even when it was right. Don't mistake that for being convinced.
It can't check its own work. Asking "are you sure?" produces reassurance, not verification.

What about your private information?

A fair question, and one people don't ask often enough.

Assume anything you type could be read by a human reviewer or used to improve the system, unless you've specifically turned that off. Most major tools now have a setting for it, usually under Data Controls or Privacy. Find it on day one.

Don't paste these in

Passwords, bank details, ID numbers, someone else's medical records, or confidential work documents — especially the last one, where the risk is your employer's policy rather than the tool itself.

For ordinary use — a cover letter, a recipe, an awkward email — this isn't worth losing sleep over. Just know where the line is before you need it.

"Is it going to take my job?"

The honest answer is more boring than either side of this argument wants.

Not your whole job. Some tasks in it, yes — the repetitive, formatting, first-draft, tidying-up parts. The parts involving judgement, relationships, responsibility and knowing what actually matters are proving much harder to automate, partly because AI cannot be held accountable for being wrong.

What is observably happening is a split between people who use these tools and people who don't, doing the same job at noticeably different speeds.

The thing to be slightly concerned about isn't AI. It's being the last person on your team still doing the boring part by hand.

Your first twenty minutes

Don't read more articles. Do this instead. It's deliberately unambitious, because the goal is a habit, not a breakthrough.

1Open one tool. ChatGPT, Claude or Gemini. Free account, two minutes.
2Give it something real and annoying. Not a test question — an actual task you've been putting off. The email you don't want to write.
3Be specific, and say how long. Use the four things above. Tell it the facts it couldn't possibly know.
4Reject the first answer. Say what's wrong with it. Ask again. This is where most people quit, and it's the step that makes the difference.
5Check anything factual. Numbers, dates, names — anything you'd be embarrassed to get wrong.
6Do it again tomorrow. One task a day for a week. That's the whole learning curve.
If it doesn't click immediately

That's normal. Most people's first three attempts are disappointing, because they ask something vague and get something vague back. The fourth attempt — where you give real context and push back on the answer — is usually the one where it becomes obvious why people care.

The one-sentence version

Pick one free tool, give it a real task, be specific about what you want, argue with the first answer, and check anything factual before you rely on it.

That's the whole thing. Everything else being written about AI right now is either detail you'll pick up naturally, or noise designed to make you feel behind.

You're not behind. You're starting on a Tuesday instead of two years ago — and that matters considerably less than anyone selling a course would like you to believe.

When you're ready for a second tool 521 AI tools, checked by a human

No scraped junk, no dead links, no pay-to-rank. Filter by what's genuinely free — no signup, no email wall.

See the genuinely free ones →
📬 The 5 best new AI tools, every Tuesday
One short email. No spam, unsubscribe anytime.