← Home
Start here

New to AI? Start here.

You do not need to be technical. You need a real problem, an approved tool, and the habit of checking the answer. This page gets you from curious to your first useful result.

No technical skills needed Invented data only

1What these tools are good at, and bad at

An AI assistant is a program that writes and answers in plain language. It is very useful, and it has clear weak spots. Knowing both makes you a safer user.

Good at

  • Drafting and redrafting
  • Summarising long documents
  • Explaining something in plainer words
  • Turning a rough process into clear steps
  • Writing small tools when you describe what you want
  • Asking you questions to sharpen your thinking

Bad at

  • Arithmetic and counting (use a spreadsheet or code for numbers)
  • Knowing your organisation's rules unless you give them
  • Knowing what is true today
  • Admitting it does not know, unless you ask it to
  • Consistency, unless you give it a method

That is why every use case on this site either uses ordinary code for the numbers or gives the AI a written method to follow.

2Five words you will keep hearing

You will see these on this site and in meetings. Each one is simpler than it sounds.

Prompt
What you type to the AI. A good prompt says who you are, what you want and what a good answer looks like.
Vibe coding
Describing a tool in plain English and letting the AI write the code. You test it and ask for changes. See it in the pay gap analyser.
Skill
A saved set of instructions the AI follows every time, so the same question gets the same method. See it in job evaluation.
Agent
An assistant set up in advance with standing instructions and its own documents. See it in the remuneration team assistant.
Grounded
Answering from the documents you gave it, not from general knowledge. A grounded assistant should point to where in your document the answer came from.

3Your first 30 minutes

Four short exercises. Use any capable AI assistant your employer allows, for example Microsoft Copilot, Claude or ChatGPT. Copy each prompt, paste it in, and read what comes back.

Invented data only, for now. Nothing in these exercises needs a real person's details. Keep it that way until you know your tool is approved for them.

Step 1: Ask it to interview you

Let the AI do the asking. This helps you find a good first task, and shows how a conversation with it works. Answer in a sentence or two. Describe the task in general terms, with no names or pay figures.

I work in HR or reward in Australia and I am new to using AI. I want to find a good first task for you to help with. Please interview me. Ask me five questions about a task I do every month that I find tedious. Ask one question at a time and wait for my answer before you ask the next. Do not ask for names, pay figures or any information about real people. After the fifth answer, suggest where you could help with that task, and where you could not. Be honest about the limits.

Step 2: Get a first draft, then push back

The first answer is a starting point. The skill is in the follow-ups. Send the first prompt, then the two that follow it, one at a time.

I work in HR in Australia. Please draft one paragraph for an all-staff email that explains the difference between a gender pay gap and equal pay. Use plain English, a warm and factual tone, and no jargon. The readers are staff across the organisation, most of whom are not HR specialists.
Make it shorter. Keep the key point and cut everything else.
Now tell me what a sceptical reader would challenge in this paragraph. Where might someone say it is unclear, one-sided or missing something?

Step 3: Make it check itself

AI assistants can sound sure when they should not. Ask them to show you where the weak points are. Send this straight after step 2, in the same conversation.

Look back at your last answer. List what you assumed, and what you are unsure of. Tell me which statements I should check against a trusted source before I use this, and say where I could check each one.

Step 4: Try a tiny tool

This is a taste of vibe coding. The AI writes a small tool, and you check it by hand. The salaries below are invented. Save the file it gives you with a name ending in .html, then double-click it to open it in your browser.

I am not a developer. Please write me a single HTML file that I can save to my computer and open in a browser by double-clicking it. It should have a box where I can type or paste a list of salaries, one per line, and a button that shows the average (mean) and the median of those salaries. Start the box with these invented numbers: 70000, 80000, 90000, 100000, 160000. Requirements: - No external libraries, links or images. It must work offline and must not send data anywhere. - Use plain English and Australian spelling. - Ignore blank lines, and tell me if any line is not a number. - Under the results, add one short sentence explaining the difference between the average and the median. Then tell me how I can check the answers by hand, step by step, using the invented numbers above.

Now do the check. Add the five numbers up and divide by five for the average. Put them in order and take the middle one for the median. If the tool disagrees with you, it is wrong. Tell the AI what you expected and ask it to fix the tool.

4Six habits that make the difference

These matter more than any clever prompt.

  1. Give contextSay who you are, who the answer is for, and what good looks like.
  2. Give it a method, not just a requestTell it the steps to follow. A method gives you the same approach each time.
  3. Ask it to ask you questions firstIt often needs more than you thought to say. Let it ask before it answers.
  4. Check every number yourselfUse a spreadsheet, a calculator or hand arithmetic. Never trust a figure because it was written confidently.
  5. Keep your promptsThey are your working. Saved prompts let you repeat, improve and explain what you did.
  6. Protect people's detailsNever paste real people's details into a tool that is not approved for them. See which tool for which data.

5Getting IT and your privacy officer on side

They are usually allies, not obstacles. Their job is to protect people's information, and yours is too.

  • Go early. Talk to them before you build or buy anything, not after.
  • Bring a specific use case. Say exactly what data it touches, for example "employee ID, gender and base salary", not "some HR data".
  • Ask what is already approved before you ask for something new. You may already have a tool you can use.
  • Offer to pilot with invented data. It lets everyone see how the tool works without any risk to real people.
  • Write down what you agreed. A short email after the conversation is enough.

For more, read the privacy primer. If you are thinking about a build that touches real pay data, the IT brief in use case 06 is written to hand to your IT team.

6Where to go next

Pick the one that matches what you want to do.