Let’s start with something reassuring. Almost nobody understands this stuff as well as they pretend to, and the people making you feel behind are usually two chapters ahead at most.

You do not need to understand how it works to use it well, any more than you need to understand internal combustion to drive. What you need is an accurate mental model — a sense of what kind of thing it is — because that’s what tells you when to trust it and when to check.

What it actually is (the useful version)

Forget robots. Forget consciousness. Here is the model that will serve you best:

It is an extraordinarily well-read assistant with a brilliant instinct for language, no memory of yesterday, no ability to check anything, and absolutely no idea what it doesn’t know.

Unpack that and everything else follows.

“Extraordinarily well-read”

These systems were trained on an enormous amount of written material. That’s why one can discuss plumbing regulations, Renaissance poetry and Excel formulas in the same conversation. It has, in a loose sense, read about all of them.

“A brilliant instinct for language”

At its core, the thing it does is predict what should come next. That sounds unimpressive until you realise how much of human work is that: drafting, summarising, rephrasing, structuring, explaining, translating between levels of expertise. It is genuinely superb at all of it.

“No memory of yesterday”

Unless a tool has explicit memory features turned on, each conversation starts from nothing. It doesn’t know you, your business, or what you told it last week. This is why giving context every time matters so much.

“No ability to check”

Unless it’s actively searching the web, it isn’t looking anything up. It’s producing what a plausible answer looks like. Usually that’s right, because plausible and correct overlap heavily. Sometimes it isn’t — and it will be exactly as confident either way.

What it’s genuinely brilliant at

Not a marketing list — the things that reliably work.

• Getting you off a blank page. A mediocre first draft in ten seconds beats a perfect one you never started. Write badly, then ask it to fix the badness.
• Making something shorter. Give it three rambling paragraphs and ask for one clear one. It will almost always improve it.
• Explaining at your level. “Explain this contract clause as if I’m smart but have never read a contract.” This use alone is worth learning the tool for.
• Changing the register. Turning your blunt note into a polite email, or a formal letter into plain language.
• Structuring a mess. Paste in twenty scattered thoughts and ask it to group them into themes.
• Being a thinking partner. “Here’s my plan. What am I not seeing? Argue the opposite case.”
• Translating between worlds. Between languages, and between jargon and English.
• Patient repetition. You can ask it the same thing five different ways and it will never sigh.

Notice the pattern: it’s excellent when you supply the truth and it supplies the shape. It gets shakier the other way round.

Where it will let you down

• Specific facts, figures and dates. Prices, statistics, legal thresholds, opening hours. It may produce something that looks exactly right and isn’t. Always verify anything you’d be embarrassed to get wrong.
• Anything recent. Its knowledge has a cut-off, and a great deal changes after it. If a tool can search the web, ask it to — and check the links it gives you actually exist.
• Citations and references. It can invent plausible-looking sources. Click every link before you rely on one.
• Arithmetic in prose. It can reason about maths but will sometimes slip on a calculation buried in a paragraph. Check the numbers.
• Knowing your specifics. Your local regulations, your client’s history, your actual costs. It will happily guess unless you tell it.
• Saying “I don’t know”. It’s getting better at this, but the default instinct is to produce an answer. Explicitly giving it permission to say it doesn’t know helps a lot.

The one skill: learning to ask

Here is the difference between people who find these tools life-changing and people who find them disappointing. It isn’t which tool. It isn’t intelligence. It’s how much context they give.

Most people type a short question, get a generic answer, and conclude the technology is overrated. Which is fair — a generic question deserves a generic answer.

Four things turn a bad prompt into a good one:

1. Who you are. “I run a two-person landscaping business.” Everything downstream changes.
2. What you’re actually trying to achieve. Not the task — the outcome. “I want this client to feel looked after, and to pay.”
3. What good looks like. Length, tone, format. “Under 150 words, warm but not chummy, no bullet points.”
4. Permission to push back. “Before you answer, ask me anything you need to know.”

That last one is the trick almost nobody uses, and it’s the highest-leverage sentence in this whole article. Watch the difference:

WEAK
"Write an email to a client about a late payment."

STRONG
"I run a small landscaping business. A regular client I like is
21 days late on a R4,800 invoice. I want to be paid without
damaging the relationship — they've been good to me for years.
Write a short email, warm, under 120 words, no legal language,
signed off as Jordan. Before you write it, ask me anything you
need to know."

The first gets you a template. The second gets you something you can actually send.

Ten prompts to use today

Copy these, change the bracketed bits, use them. Every one solves a real problem.

1. Understand anything. Explain [topic] to me as if I'm intelligent but have never encountered it. Use an everyday analogy. Then tell me the one thing most people get wrong about it.
2. Escape a blank page. I need to write [thing]. Ask me five questions that will give you what you need, then write a first draft.
3. Difficult message. I need to say [hard thing] to [person] without damaging the relationship. Here's the context: [context]. Give me three versions: gentle, direct, and firm.
4. Decide something. I'm choosing between [A] and [B]. My situation: [context]. What matters most: [priorities]. Argue for each properly, then tell me what you'd do and why.
5. Make it shorter. Cut this to half the length without losing anything important. Keep my voice. [paste]
6. Understand a document. Summarise this in plain English. Then list anything I should be worried about, and anything that seems unusual. [paste]
7. Prepare for a conversation. I have [meeting] about [subject]. Give me the five questions I should ask, and the three I should be ready to be asked.
8. Learn faster. Teach me [skill] in a way that gets me doing something useful in the first twenty minutes. Assume no background. Then give me a practice task.
9. Find your blind spot. Here's my plan: [plan]. Be genuinely critical. What's the most likely reason this fails? Don't be diplomatic about it.
10. Explain it to family. Rewrite this so my [parent/teenager] would understand and actually read it. Same information, no condescension.

What never to put in

The convenience is real and so is the risk, and hardly anyone explains the second part properly. Depending on the tool and your settings, what you type may be stored, and may be reviewed by humans for quality or used to improve the system.

So, a simple rule: don’t paste anything you’d be uncomfortable seeing on a screen in someone else’s office. Concretely:

• ID numbers, passport numbers, tax numbers
• Full banking or card details
• Passwords, of any kind, ever
• Other people’s medical information
• Client data you’re contractually obliged to protect
• Anything covered by a confidentiality agreement

You can almost always get the same help by anonymising first. “A client” instead of a name. “R4,800” instead of an account. The advice is identical; the exposure isn’t.

It’s also worth spending two minutes in your tool’s settings. Most now let you turn off training on your conversations, and many offer a temporary chat mode that isn’t saved.

The honest bit about jobs

You’ve heard the two extremes: everything is fine, and everything is over. Neither is useful.

What’s more defensible: these tools are very good at tasks and much weaker at jobs. A job is a bundle of tasks plus judgement, relationships, accountability and physical presence. Some of the tasks in your bundle are now much cheaper to do. That changes what your day looks like more than whether you have one.

The realistic near-term risk isn’t a machine replacing you. It’s competing against someone in your trade who has learned to do in one hour what still takes you four. That gap is closeable, this week, by anyone — which is the entire reason this site exists.

And there’s a genuinely optimistic side that gets far less airtime. The person who most benefits from an infinitely patient explainer is not the specialist. It’s the person who left school early, or works in a second language, or never had anyone in their family who could explain a contract. This technology is unusually good at levelling that particular field.

Your first week

Don’t try to learn it. Use it for real things, small ones, until it becomes a reflex.

• Day 1. Pick one tool. Any of the major ones. Stay with it — switching constantly is why people never get good.
• Day 2. Use prompt 1 on something you’ve genuinely never understood.
• Day 3. Use prompt 3 on a message you’ve been avoiding sending.
• Day 4. Paste in something long and boring. Ask for the plain-English summary.
• Day 5. Use prompt 9 on a real decision. Let it be rude to you.
• Day 6. Deliberately catch it being wrong about something you know well. This builds the right instinct better than any warning.
• Day 7. Notice which of the six felt useful. Do more of that one.

That’s the whole on-ramp. Not a course, not a certification — seven small real uses. By the end of it you will have a working sense of what this thing is for, which is more than most people who talk about it confidently have.

You’re not behind. You just hadn’t been shown. Now you have.