Your first useful conversation, step by step
Get one genuinely useful result today, using a real message you need to write.
The best way to learn is to use AI for something you actually need to do today. Writing a message is perfect.
Step 1. Think of a real message. A reply to a neighbour, an email to your child's school, a note to a colleague, a complaint to a company.
Step 2. Tell the assistant four things:
- who it's to
- what it's about
- the tone you want (friendly, firm, polite)
- your rough notes, in your own words. Spelling mistakes are fine.
Step 3. Read what it writes. It won't be perfect first time, and that's normal.
Step 4. Ask for changes, in plain words. "Shorter." "Less formal." "Don't apologise so much." "Mention that I'm free on Thursday." Keep going until it sounds like you.
Step 5. Check and send. Fix any details it got wrong and send it yourself.
That's the whole skill: tell it what you want, then keep chatting until it's right. Congratulations, you've just used AI for real.
Why start with a real task
People who try AI with a vague test ("tell me a joke") often shrug and never come back. People who use it on a real job they were putting off usually come back the next day. So let's do a real job.
A worked example
Say you need to email your landlord about a boiler that keeps breaking down.
Your first message to the assistant:
I need to email my landlord about the boiler. It's broken down three times since October, the engineer said it needs replacing, and I've been without hot water for two days. I want to be polite but firm and ask for a date when it will be replaced. Please write a short email.
Notice the four ingredients: who (landlord), what (boiler, the facts), tone (polite but firm), what you want back (a short email).
What happens next. You'll get a draft in seconds. Read it. Maybe it's a bit long, or sounds stiffer than you'd like.
Ask for changes in plain words:
Make it shorter and a bit warmer. Add that I'm happy to give access any weekday morning.
Then:
Perfect. Can you give me a subject line too?
Five things that help
- Give the facts. The assistant only knows what you tell it. Names, dates and details make it far better.
- Say who's reading. "My manager", "a 10-year-old", "a company's complaints team" all change the result.
- Don't worry about wording. Write how you'd explain it to a friend. Typos are fine.
- Ask for options. "Give me three versions: one formal, one friendly, one very short."
- Keep the conversation going. Changes are the point, not a sign it failed.
Before you use it
Always read the final version. Check that names, dates and numbers are right, and that it hasn't added anything you didn't say (for example a promise, or a deadline you didn't mention). Then send it yourself.
Other good first tasks
- Turn a long work email into three bullet points.
- Plan meals for the week from what's in your fridge.
- Explain a confusing letter in simple words.
- Write a birthday message for a card.
You'll find step-by-step versions of these in our task guides.
Why this approach works
AI assistants produce the most likely helpful response to what they're given. A short, vague request ("write an email to my landlord") leaves the model to guess the facts, the tone and the length, so you get something generic. A request with context removes the guesswork. This is the core idea behind what people call prompting (lesson 7 covers it in depth).
Equally important is the back-and-forth. Assistants keep the whole conversation in view, so every follow-up builds on what came before. Experienced users rarely accept the first answer; they treat it as a draft to shape.
A fuller worked example
Situation: you're disputing a parking charge.
Opening request (with context):
I got a £100 parking charge from a private car park on 3 November. I paid for 2 hours using the app at 10:02 and left at 11:48. I have the payment confirmation screenshot. The letter says I didn't pay. I want to appeal politely and clearly, attaching the evidence. Please draft the appeal, under 200 words, and list what I should attach.
Typical follow-ups:
- "Make the first sentence say clearly that I'm appealing."
- "Remove the bit about it being stressful; keep it factual."
- "What's the deadline usually for appeals like this in the UK, and what happens if they reject it?"
The last one is a different kind of question: a factual one. The assistant may answer well, but this is where you should check with an official source (in this example, the parking company's own letter and the independent appeals service it names). Writing help and factual answers deserve different levels of trust.
Getting it to sound like you
Assistants tend to write in a polished, slightly formal style. To make it sound more like you:
- Paste a message you've written before and say "match this style".
- Ask it to "use short sentences and everyday words".
- Tell it what to avoid: "no exclamation marks", "don't start with 'I hope this finds you well'".
Checking the result properly
Before you send anything an assistant wrote, check for:
- Invented details. Assistants sometimes fill gaps with plausible inventions, such as a reference number or a date.
- Promises you didn't make. "I'll call you tomorrow" may appear when you never said it.
- Tone in context. It doesn't know your relationship with the reader.
- Facts and figures. Anything it stated as a fact (a law, a deadline, a price) should be checked.
Building the habit
The fastest way to improve is to use AI for one small real task a day for a week. You'll quickly learn what it's good at for you. Making it a habit gives you more ideas when you're ready.
Try this prompt
I need to write a message to [who] about [what]. I want it to sound [friendly / firm / polite]. Here's what I want to say in my own words: [your rough notes]. Please write a short version I can send.
Replace the [brackets] with your own details
On haystak
Quick check · 1 of 3
Which request will get the better result?
Next step
What AI gets wrong, and how to spot it
Know the mistakes AI makes most often, so you can catch them before they cause problems.