1. Your First 10 Minutes With AI

By the end of this lesson you will have sent your first three prompts and gotten something useful back. It takes about 10 minutes and costs nothing.

Step 1: get an account

1. Go to claude.ai and sign up. The free tier is enough for this entire course.
2. You land on a screen with one text box. That is the whole interface. There is nothing else to learn.

ChatGPT, Gemini and Copilot all work the same way, and everything in this course transfers. We use Claude because the free tier handles documents well.

Step 2: your first prompt

Do not start with "hello" or "can you help me". It works, but it wastes a turn. Type the actual task.

Copy this, changing the topic to something you actually deal with:

Explain how car insurance excess works, to someone
who has never bought insurance before. Use a real
example with numbers.

Notice the three parts, because every good prompt has them:

• The task: explain how car insurance excess works
• Who it is for: someone who has never bought insurance
• The shape of the answer: a real example with numbers

Step 3: the follow-up that most people never send

You are in a conversation, not a search box. It remembers what you just asked. So the second message is where the value is:

Now give me the 3 questions I should ask before
choosing an excess amount.

You did not repeat the topic. It knows. This is the single biggest difference from a search engine, and new users miss it because a search box has no memory.

Step 4: three real tasks to try right now

Pick whichever is closest to something on your actual to-do list. Copy it, change the details, send it.

Understand something confusing:

I got this message from my bank and I do not
understand it. Explain what it means and what,
if anything, I need to do:

[paste the message]

Draft something you have been avoiding:

Write a short message declining a wedding invitation
from a close friend. I genuinely cannot go because
of work. Warm, not over-apologetic, under 100 words.

Make a decision faster:

I am choosing between a 1 year and a 2 year phone
contract. List the tradeoffs in a table, and tell me
what information I would need to decide properly.

That last instruction, "tell me what information I would need", is worth stealing. It turns a confident guess into a checklist, and it works on almost any decision.

What one of those actually returns

We ran the phone contract prompt exactly as written above. Here is a real extract of what came back, not a mock-up:

Factor	1 year	2 year
Monthly cost	Usually higher	Usually lower, subsidised phone deal
Flexibility	Switch sooner	Locked in, early exit fees
Total cost	Higher monthly x 12	Lower monthly x 24, often lower overall

Then, because we asked what information we would need, it listed 6 things: whether you are happy with your carrier, how long you keep phones, the exact monthly difference, the early termination fee, whether you need current hardware, and coverage where you actually go.

It finished with something we did not ask for and which is the most useful line in the answer:

If the 2-year monthly savings exceeds the exit fee, the 2-year contract is cheaper unless you're 80%+ confident you'll switch carriers within the year.

Look at what happened there. It did not tell you which contract to buy, because it cannot know. It told you the rule that decides it and the 2 numbers you need to apply that rule.

That is the shape of a genuinely useful AI answer, and it came from 1 extra clause in the prompt. Not "which should I pick", but "what information would I need to decide properly".

When the answer is wrong or unhelpful

It will happen in your first ten minutes. Three fixes, in the order to try them:

1. Say what is wrong, do not start over. "Too formal." "You missed the deposit." "Shorter, half that length." It keeps everything else and changes that.
2. Give it the thing it is guessing about. If it invents details about your situation, paste in the real ones. Guessing is what it does when you leave a gap.
3. Start a new chat if it has gone off the rails. A long conversation that took a wrong turn tends to stay wrong, because everything it already said is still in front of it.

The third prompt, and what it returned

The explaining prompt is the one people underuse, so here is a real one and the real answer. We asked:

Explain what a VPN actually does, in plain English.
I am not technical. Keep it under 150 words and tell
me the main thing people get wrong about them.

It came back at 154 words. It described a VPN as an encrypted tunnel that routes your traffic through a middleman server, then split the answer into what it protects and what it does not, which we did not ask for and which is the right structure for the question. Under "what it doesn't do" it listed making you anonymous, hiding you from law enforcement, and protecting you from viruses or phishing.

Then the part we did ask for:

People think VPNs make them untraceable online. They don't. A VPN just shifts trust from your ISP to the VPN provider.

That is a genuinely good answer to a question a lot of people have quietly wondered about for years. It took 20 seconds and one sentence of prompt.

The clause doing the work is "tell me the main thing people get wrong". Without it you get a definition. With it you get the misconception, which is usually the reason you were confused in the first place.

Which app, and what it costs

You do not need to pick carefully. The free tier of any of the main ones is enough for everything in this course, and switching later costs you nothing because there is nothing to migrate.

What actually differs between free and paid, in practice:

	Free tier	Paid, roughly $20/mo
Good enough to learn on	Yes	Yes
Messages per day	Capped, and you will hit it	Much higher
Model quality	Usually the smaller model	The better one
Long documents	Limited	Handles real files

Our advice: use the free tier for 2 weeks. If you hit the daily cap more than twice, that is your answer. If you never hit it, you were never going to get $20 of value out of it.

Five things people get wrong in the first hour

1. Typing a search query instead of a request. "best laptop 2026" gets you a shrug. "I edit photos, my budget is $1,200, I need battery life over raw speed, what should I look at and why" gets you an answer.
2. Accepting the first response. The first answer is a draft. Almost every good result in this course took 2 or 3 rounds.
3. Not saying who they are. One sentence of context about your situation changes the whole output. It is the cheapest improvement available.
4. Believing a number because the rest was right. Fluency is not accuracy. Lesson 5 has a real example of this and it is worse than you would guess.
5. Being polite at the expense of being clear. "Could you possibly help me with maybe writing something about" wastes the part of the message that should carry your facts. Be direct. It has no feelings to spare.

What to expect, honestly

It is reliably good at	It is unreliable at
Explaining things at any level	Current prices and dates
Drafting text you will edit	Anything that happened recently
Reformatting and summarising	Specific legal or medical advice
Working with text you paste in	Arithmetic buried in prose

The pattern behind that table: it is strongest when working with what you gave it, weakest when recalling facts from memory. Lesson 8 covers checking its work. For now, use it on the left column.

Before you move on

Actually send three prompts. Reading this lesson teaches you nothing; the muscle memory is the point, and it takes about 10 minutes.

Next

Lesson 2 is the one that changes your results most. We send the same request twice, once badly and once well, and show you both real outputs side by side.