Hi, I'm Mariia, and I designed this course. A quick note first: the presenter you're watching is an AI avatar of me, created with my consent, so I can keep these videos up to date as AI tools change.

Over the next five days, you'll go from one-line requests to prompts you can reuse at work. Each day takes about half an hour: a short video, a worked example, and a real task in your own AI assistant.

This isn't a list of tips. Every day, you'll practise on a real task, and by Day 5 you'll build a prompt template for something you do every week.

Before we start: what do you use AI for most? Pick one below, and let's begin.

You ask an AI assistant to write an email. It gives you something polite, correct… and useless. You rewrite half of it yourself. Today we'll see why that happens, and it's not because the tool is bad.

Here's the prompt most of us write. I ran it three times. Notice anything? Every version is fine. None of them is yours. None mentions the real reason, the new date, or who's reading.

An AI assistant doesn't know your project. It writes the most likely answer to what you asked. When you leave something out, it fills the gap with the average: the email that fits every project, which means it fits none.

So a vague prompt isn't a small problem. Every gap you leave is a decision the model makes for you, and it will always choose the safest, blandest option.

Now watch what happens when we fill those gaps. The task: tell the client the launch moves by two weeks. The context: a supplier delay, and a client who values honesty. The constraints: under 120 words, calm tone. The format: a subject line plus three short paragraphs.

Same tool, same model. The difference is the brief. And that's the key idea for this whole course: a prompt is a brief. If a new colleague couldn't do the task from your message, neither can the model.

In the next lesson you'll get the four elements as a checklist you can keep. Then it's your turn.

Yesterday you learned that a prompt is a brief. Today we'll add the most powerful thing you can put in a brief: your own material.

Ask for a "professional" email and you'll get the model's idea of professional: formal, long, a little stiff. Your idea might be warm and short. An adjective can't tell the difference. An example can.

Instead of describing the style, show it. Paste a message you've already sent, and say: match this tone. One or two good examples do more than a paragraph of adjectives.

Your material doesn't have to be an example of style. It can be the notes you want summarised, a report, or even a photo of a whiteboard. Most assistants accept all three.

Two habits make material work. First, label it: tell the model what it is, and mark where it starts and ends. Three quotation marks work fine. Second, say what to do with it: summarise it, copy its structure, or match its tone.

And one check before you paste: documents are full of names and numbers. Remove what the model doesn't need. You learned how yesterday.

Let's see it on a real example: messy meeting notes, and a manager with two minutes to read them.

Yesterday you saw the AI copy a format perfectly and still get the facts wrong. Today is about what to do when the answer is wrong.

An AI assistant sounds just as sure when it's wrong as when it's right. It added a currency to our budget, then changed it. It invented a date. None of it was flagged. Confident is not the same as correct.

When an answer misses, there's usually one of four causes. The model didn't have the facts. The limits were vague. It didn't know what good looks like. Or you asked for too much at once.

Each cause has its own fix. Add the facts. Name the limit: length, tone, audience. Show an example. Or split the task into steps. Diagnose first, then fix. Just pressing "regenerate" is a lottery.

Before you use any answer, run a three-question check. Is every fact in my material, or in a source I trust? Are names, numbers and dates exactly right? And would I sign my name under it?

One more thing to check: bias. Ask for a job ad for a "strong, young team player", and the model will describe exactly that person, and leave everyone else out. The words in your prompt shape who appears in the answer.

Let's practise on answers that look fine, and aren't.

Some jobs are too big for one prompt. Today you'll learn to split them into steps, and to get answers in a shape you can use straight away.

One job, one prompt. Ask for an agenda, a materials list, a budget and an invitation all at once, and the model does each of them a little. Something gets skipped, something gets rushed, and you can't tell which part went wrong.

Instead, build a chain. Step one: the agenda. Step two: the materials list, based on that agenda. Step three: the invitation, based on both. Each step uses the one before it.

Stay in the same chat, so the model remembers each step. And check between steps: a mistake in step one travels all the way to step three.

Ask for the shape you need. An agenda works best as a table: time, session, goal. Materials as a checklist. Name the columns, and the model will fill them.

So: one job per prompt, a check after each step, and a clear shape for every answer. That's how long tasks stay under control.

Let's plan a real workshop, one step at a time.

Welcome to the last day. You've learned to write a clear request, to work from your own material, to check the answer, and to break big jobs into steps. Today you'll turn all of that into something you keep: a prompt template.

A template is a prompt you write once and reuse every week. The parts that stay the same are written out: the structure, the length, the tone, the rules. The parts that change are placeholders in square brackets: the audience, this week's notes.

Add one line that protects you: "Use only facts from the notes. If something is unclear, list it under To check instead of guessing." It helps. It doesn't do your checking for you.

Before you trust a template, test it on two different real cases. If it works on both, keep it. If it doesn't, fix the template, not just the answer.

Build yours today, check it against four criteria, and take the final check for your certificate.