To start using AI, open a free chat tool such as ChatGPT, Claude or Gemini, type a real request in ordinary language with a little background, and ask for changes until the reply is useful. You don't need to install anything or learn special commands. If you can write a message to a colleague, you can use AI.
What takes practice is knowing what to ask for and checking what comes back. This guide assumes you've never touched it: by the end you'll know what to try first, what to ignore and where the limits are.
What is AI, in plain words?
The AI most people mean is a chat tool that writes a fresh reply to whatever you type. You open a web page or an app, type a message, and it answers. The answer is written on the spot, based on patterns the system learned from an enormous amount of text, rather than copied from one source or pulled from a database of stock replies.
That has two consequences worth knowing from day one. It can produce something useful on almost any topic, including your own work, in seconds. It can also produce something that sounds confident and is wrong. Both are normal: the tool generates the most plausible text it can, and plausible and correct are different things.
Treat it like a capable colleague who has read a great deal but hasn't seen your files and doesn't know your business. It's good for a first draft and unreliable as a source of facts.
What should you type?
A real request with context
Vague: "help me with my shop".
Useful: "I run a small shop and I need to write a message to my customers explaining that we'll close for three days for renovation. Keep it short and friendly."
The second one works because it says who you are, what you need and the tone you want. Giving that context is most of the technique; there are no commands to memorise.
The material itself
If you want help replying to a message, paste the message. If you want a summary of a document, paste the document. If you want a project plan, describe the project properly.
Most disappointing results come from describing something instead of showing it. A chat tool can't see your screen or your inbox; it only knows what you put in front of it. Before you paste anything about clients or colleagues, read what happens to your data.
A request for changes
The first reply is rarely the one you use. Answer it: "shorter", "more formal", "remove the last paragraph", "make it sound less like a sales pitch". This back and forth is normal, and it's where most of the value comes from.
People who get little out of AI tend to try once, dislike the result and stop. People who get a lot out of it type three more sentences.
What should you try first?
Pick one real task you'd do anyway. Ten minutes is plenty, and AI in ten minutes gives you a timed routine if you'd like one.
Write something you've been putting off. A difficult email, a polite refusal, a request for payment, an apology for a delay. Describe the situation in three or four sentences, ask for a draft, then edit it until it sounds like you.
Turn messy notes into order. Paste a jumble from your phone or a notebook and ask for it grouped by theme, or turned into a short plan with next steps.
Get something explained. Paste a paragraph from a contract, a letter from your energy supplier or a technical description, and ask for it in plain language. Ask it to point out anything that's unclear.
All three work in a free chat tool with no setup.
What can't AI do?
A chat tool can't see what you haven't given it, can't be trusted on facts without checking, and doesn't carry out tasks for you. This matters more than anything else on the page, because it's where people get caught out.
It can't see your files, inbox or calendar unless you paste them in or connect them. Ask about "the email from the supplier" and it has no idea which email you mean.
Its memory of you is limited. Some chat tools now keep notes between conversations, including ChatGPT and Claude, but what they keep depends on the tool, your plan and your settings. Don't assume it remembers last week's conversation in detail; if something matters, paste it again.
It can't reliably check facts. It can be out of date, and it can invent a detail that fits the shape of the sentence. The providers say so themselves: Anthropic's help centre warns that Claude can give incorrect or misleading answers. If a number, date, law or name matters, confirm it somewhere else.
A basic chat doesn't take actions. It writes text. It won't send the email, update the spreadsheet, book the meeting or file the document. You still do those, or you use something built to do them, which is where agents come in.
Where do AI agents fit in?
A chat tool is a conversation: you ask, it answers, you act. An agent is an AI given a job, the tools to do it and rules about when to stop and ask. It can remember how you like things done, run tasks on a schedule and follow up without being prompted, so the work itself moves off your desk along with the wording. If you're curious how that remembering works, see how AI agent memory works.
If you've never used AI, leave agents for later. Get comfortable asking good questions and judging the answers first, because that skill carries straight over. When you want the full comparison, read AI usage vs an AI agent.
A gentle first week
- Day 1: write one thing you've been avoiding.
- Day 2: paste a messy set of notes and ask for order.
- Day 3: ask it to explain something from your work in plain language.
- Day 4: look back at the best result so far and work out what made it good. Was it the context you gave, the tone you asked for or the revision you requested? That's the start of a habit you can repeat.
- Days 5 to 7: use it only when it saves you real time, and skip it when it doesn't. Nobody gives prizes for using AI more often.
When should you skip AI?
If the task is a two-line message you could write in the time it takes to describe it, write it yourself.
If the answer has to be exactly right and you have no way to check it, start with a source you trust instead.
If someone is telling you to adopt AI urgently, be sceptical. Nothing about the tool demands urgency, and what works is small, regular use tied to work you already do.
Common questions
Do I need to pay to start using AI?
No. ChatGPT, Claude and Gemini all offer a free version that covers everything in this guide. Paid plans mainly add higher usage limits and stronger models, so try the free version on a few real tasks and only pay once you know where it holds you back.
Do I need technical skills to use AI?
No. If you can write a clear message to a colleague, you have the skill that matters, and there's nothing to install beyond an app or a web page. Good results come from practice with real tasks rather than from training.
Is it safe to type personal information into an AI chat?
Be careful with it. What you type goes to the provider's servers and is handled under their terms, which may allow them to use it to improve their models unless you change a setting. Avoid pasting passwords, bank details, health records or other people's private information, and if it's for work, check what your employer allows first.
Which AI tool should a beginner use?
Any of the major free chat tools is a fine place to start, and the differences between them matter far less than how you use them. Pick the one that's easiest for you to open, use it for a week on real tasks, and judge it on whether it saved you time. You can try another later with the same requests and compare.