AI, plainly explained

What AI is (and isn't)

When people say AI today, they usually mean tools like ChatGPT, Claude or Gemini. You type something, and it writes something back. That's the core of it.

Under the hood, these tools have read an enormous amount of text. From that, they learned which words tend to follow which other words. When you ask a question, they predict a likely answer, one word at a time.

That sounds simple. It is, in a way. But it turns out that predicting text very well gets you surprisingly far. It can write an email, summarise a long document, or turn messy notes into a tidy list.

What it isn't: a person. It doesn't know you, your clients or your prices unless you tell it. It doesn't understand things the way you do. And it doesn't check whether what it says is true.

Think of it as a very fast, very well-read intern. Helpful. Eager. But you'd never let it send something important without reading it first.

What it's good at

AI is strongest at tasks with words, patterns and repetition. Exactly the kind of work that fills up a freelancer's week.

Writing a first draft. A reply to a client who asks the same thing as the last five clients. A quote based on three lines of notes. A LinkedIn post from a voice memo.

Summarising. A long email thread, a contract, meeting notes. It pulls out the main points in seconds.

Reformatting. Turning a messy list into a table. Turning your notes into your quote template. Turning a long text into a short one.

Sorting and labelling. Which emails are questions, which are invoices, which can wait. It's not perfect, but it's good enough to save real time.

The common thread: it does the boring first 80%. You do the last 20% that needs your judgement.

Where it goes wrong

AI makes mistakes. Not sometimes – regularly. And it makes them with full confidence.

It can invent facts. Ask it for a source and it may give you one that doesn't exist. Ask it for a number and it may simply guess. This is often called 'hallucinating'.

It doesn't know your context. It doesn't know that this client always pays late, or that you agreed on a discount last month. Unless you tell it, it fills the gaps with something plausible.

It's bad at precise maths and counting. For invoices, the numbers should come from your bookkeeping software, not from AI.

And there's privacy. What you type into an AI tool goes to a company's servers. Some tools use that to train their models. So think before you paste in client data, and check the settings.

None of this means you shouldn't use it. It means you should use it for drafts and suggestions, and keep yourself in the loop.

What it means for a freelancer

You don't have an IT department. You don't have an assistant. Every hour spent on admin is an hour you can't bill or spend on work you enjoy.

That's where AI can genuinely help. Not by replacing what you do, but by taking over the bits around it. The inbox. The quotes. The follow-ups. The newsletter you keep postponing.

Start small. Pick one task that annoys you every week. Try handing the first draft to AI. Check the result. If it saves you time, keep it. If not, drop it.

Be realistic. A good setup might save you a few hours a week. That's not a revolution. But a few hours, every week, adds up to a lot.

And stay in charge. You decide what goes out, under your name. AI drafts. You decide.