Start here with AI at work

You do not need to understand every model, buy three subscriptions or turn your business into an experiment. Start with a small job you already understand well enough to check.

A clear set of collection instructions is a sensible first attempt. Deciding which employee to dismiss is not.

This path takes you from a blank chat box to one useful, reviewed piece of work. Follow it in order if you are starting from scratch. If you already have an approved account, go straight to the prompt guide.

What you'll make

a short draft, a record of the checks you made and a prompt you can use again.

What you'll need

use fictional material until you know which account and information your organisation permits. Every practice example in this learning path is fictional. None is presented as a captured product response.

1. Pick a job with a finish line

Choose something small enough to judge. Good first tasks include:

• Turning a rough process into a checklist
• Rewriting a public product explanation in plainer English
• Sorting non-sensitive notes into questions to resolve
• Comparing two short, approved documents for differences

Write down what a useful result would look like before you ask for it. “Make this better” leaves quite a lot of the job to chance.

Try this brief in your own notes:

The job: Turn these approved instructions into a clear checklist.
The reader: Someone doing this for the first time.
The result: No more than eight steps, in the right order.
The facts that must survive: Every time, quantity and exception.
The check: Compare every step with the original before using it.

If you cannot describe how you will check the result, choose an easier first task.

2. Choose one suitable tool

Start with the tool your organisation has approved and already pays for. For this learning path, an ordinary text conversation is enough.

If you have a genuine choice between ChatGPT and Claude, use the short comparison exercise. It helps you judge a tool against your work without pretending that one app wins every job.

A free account can be enough for fictional practice. Real business information needs a separate decision about the account, plan, permissions and data involved.

3. Set up before adding work

Follow the account and workspace setup guide. Check your sign-in, privacy settings and project visibility before you upload anything.

Three words cause unnecessary confusion:

• Your account is your sign-in
• Your organisation’s workspace is the managed area you may belong to
• A project groups related work inside the app

Creating a project does not convert a personal subscription into an organisation-managed service.

You can leave connectors, browser access and automation until you have a task that actually needs them.

4. Give the tool a proper brief

Follow the first prompt walkthrough. It includes fictional notes, a complete prompt and a follow-up that fixes a specific problem.

Tell the tool what to produce, who will read it, which material to use and what must remain unchanged. Review the answer, then give a precise correction. This follows the basic prompting approach described in OpenAI’s guidance.

You do not need an elaborate persona. “You are the world’s greatest business genius” supplies very little information about the collection time.

5. Check it before it leaves your desk

Use the files and answer-checking guide before working with documents or relying on a factual answer.

For your first draft, check:

1. Did every important fact survive?
2. Did the tool add a fact, promise or deadline?
3. Does each step make sense in the real world?
4. Is the result suitable for this reader?
5. Who needs to approve it before it is used or shared?

AI can produce convincing errors, including made-up references. Open the underlying source rather than treating confident wording as evidence. OpenAI describes these limitations here.

Keep the reviewed version somewhere you control, alongside the source and reusable prompt. Your first success is a modest piece of work you can trust after checking it.

Keep this: Learn one useful task, check it properly, then repeat it.

When a word gets in the way

The plain-English glossary explains LLMs, prompts, tokens, context windows, agents, APIs and the rest. You can look up one term and get back to the job.

Sources and status

Official prompting and reliability guidance checked on 4 October 2026. This is an editorial learning path, not a product benchmark or a promise of time saved. See our editorial approach.