I guarantee you'll get better results from your chat when you finish - even if you think you already know this. Just chat. No agents, no plugins. Work through in order.
No sign-up, no email collected, nothing sold at the end. Here's why it's genuinely free.
Before you learn anything, write down one real thing you want AI to help you with. Not "learn AI" - something specific, like "write replies to customer emails" or "come up with social media captions for my shop."
Like thisThink of this like deciding what you actually need a car for before you walk onto a forecourt. A family car, a sports car, and a tow-car are all "cars," but you'd pick very differently depending on the job.
You'll use this same task as the running example all the way through, so pick something real from your actual work.
Example. My task: "I run a small landscaping business and need to write quotes that sound professional but not stiff."
Open a blank note on your phone or computer (or a blank piece of paper). Write these two sentences, filling in the gaps yourself:
Save the note. You'll copy these sentences into an AI chat in later stages.
1. Why start with a real task instead of jumping straight into tips and tricks?
Because tips only make sense once you have a real problem to try them on.
2. True or false: your task should be something broad, like "get better at AI."
False. Specific and real works far better than vague and general.
Think of a chat AI like an extremely well-read guesser. It has read a huge amount of text and learned the patterns of how words tend to follow each other. When you type something, it isn't looking anything up. It's predicting, word by word, what a good reply would probably look like.
Like thisIt's a bit like a very experienced mechanic guessing what's wrong with your car just from the sound of the engine - usually right, because they've heard thousands of engines, but still a guess, not a diagnostic readout.
Two things follow. It doesn't "know" things the way a person does, so it can sound completely confident while still being wrong. That's called a hallucination - not a glitch you can switch off, just something to always double-check. And unless a product specifically adds memory, it won't remember your last conversation once you start a new one.
If you ask "what's the weather in London right now" in a plain chat with no way to check current information, it can't actually know. It might admit that, or guess a plausible-sounding answer.
Most AI apps today have a "search the web" ability, and in many of them it's switched on by default - so it can fetch a real, current answer instead of guessing. That searching ability is a tool added on top of the core AI, not something the underlying model does by itself. It's worth knowing whether search is on or off in whatever app you're using.
Open any chat AI app (ChatGPT, Claude, Gemini - whichever you have). Type this exact question and press send:
What was the top news story yesterday?
If it says something like "Searching the web..." before answering, that's search switched on. If it answers straight away with no mention of searching, check whether your app has a search option. If it says it doesn't have access to today's news, that's it being honest.
The point isn't which one happens - it's learning to notice which mode you're in before you decide how much to trust the answer.
1. In plain terms, what is a chat AI doing when it replies to you?
Guessing the most likely next words, based on patterns it learned from huge amounts of text.
2. Why shouldn't you always trust a confident-sounding answer?
Because sounding confident and being correct are two completely different things.
3. Does chat AI remember your last conversation automatically?
No - not unless the app you're using specifically adds a memory feature.
4. If an AI app says "Searching the web..." before answering, what does that tell you?
That it's checking a real, current source rather than guessing from patterns - you can trust the answer more.
"AI" isn't one single product - it's a category. Claude, ChatGPT, Gemini and others are made by different companies, trained differently, and are genuinely better or worse at different jobs. None of them is "the best" for everything.
Like thisLike Ford, BMW, and Toyota: all cars, not identical. You wouldn't expect them to drive the same.
Which one suits you depends on what you're trying to do. Free versions all come with some kind of limit - an older version, a usage cap, or fewer features.
| Assistant | Good at | Free to try? | Worth knowing |
|---|---|---|---|
| Claude (Anthropic) | Writing, thinking through problems, longer documents | Yes, with a daily limit | Fewer add-on tools than ChatGPT |
| ChatGPT (OpenAI) | General use, the widest range of add-ons | Yes, with a daily limit | Can be less careful with nuanced writing |
| Gemini (Google) | Working with Google Docs/Gmail, understanding images | Yes, with a Google account | Quality can vary more by task |
| Perplexity | Looking things up with sources attached | Yes | Built for quick answers, not long writing |
| Meta AI | Free, built into Instagram/WhatsApp/Facebook | Yes, always free | Not really built for serious business use |
Go to two of these: claude.ai, chatgpt.com, gemini.google.com (you may need a free account - just an email, no payment). Type your Stage 0 task sentence into both, word for word. Read both replies side by side and note which one you'd actually use.
Claude is usually strongest on writing, thinking through a problem, and longer documents. Free, with a daily limit. Fewer add-on tools than ChatGPT.
ChatGPT is the general-use one, with the widest range of add-ons. Free, with a daily limit. Can be less careful with nuanced writing.
Gemini sits next to Google Docs, Gmail, and images. Free with a Google account. Quality can vary more by task.
Perplexity is built for looking things up with sources attached. Free. Built for quick answers, not long writing.
Meta AI is free, sitting in Instagram, WhatsApp and Facebook. Always free. Not really built for serious business use.
It doesn't stop at picking a company. Inside Claude, ChatGPT and Gemini you'll usually find a menu of versions - often named Lite, Flash, Mini, Pro, or numbered.
Like thisClaude, ChatGPT and Gemini are like three different car makers. Within one maker's range you might get an estate, a saloon, a cabriolet and a sports model - same badge, built for different jobs. A hatchback to the shops. A sports model for a harder, faster job. Using the sports model to nip to the corner shop, or the hatchback to tow a caravan, is a mismatch either way.
A quick "what's a good subject line" doesn't need the sports model. Reading a 20-page contract for risky clauses is worth switching up, even if it takes a bit longer or costs more.
Open the model menu (usually near the top of the chat, or in a dropdown next to where you type). Note the names. Pick the smallest/fastest one for a simple question, then the biggest/most capable one for the same question, and compare.
1. Is any one AI assistant the best for every job?
No - the right choice depends on what you're trying to do.
2. Name one thing that's different between the free versions.
Usage limits, which version of the model you get, or which features are switched off (varies by company).
3. Which assistant is built specifically for looking things up with sources?
Perplexity.
4. Why would one AI maker offer several different models instead of just one?
Because smaller/faster models suit simple tasks and cost less, while bigger models handle harder tasks but take longer or cost more - like a hatchback versus a sports model from the same maker.
5. Which model would suit a quick, simple question best?
The smallest/fastest one - no need for the most powerful (and often slowest or most expensive) option.
Every chat AI has a limit on how much of your conversation it can actually "hold in mind" at once - like a whiteboard of a fixed size. Everything you've typed, everything it's replied, and any documents you've shared all sit on that whiteboard together. Once it fills up, the oldest stuff at the top gets wiped to make room.
Like thisIt's a bit like a car that starts every journey completely fresh. Even if you drove 100 miles yesterday, the engine doesn't "remember" that trip today.
That's why, in a very long conversation, it can seem to forget something you said earlier. It's not being stubborn. That instruction has fallen off the whiteboard.
You tell it early on "always reply in a friendly, casual tone." Fifty messages later it suddenly sounds formal again - the earlier instruction has likely fallen off the whiteboard.
Open a brand new chat (click "New chat"). In your very first message, include everything important in one go:
I run a small landscaping business. I need help writing quotes. Please always keep replies friendly and under 100 words. My first quote: [describe the job].
Notice how the reply already matches your tone and length, because it had everything it needed from message one.
1. In plain terms, what's happening when AI seems to "forget" something?
The conversation has grown too long, and older parts have been pushed off to make room for new parts.
2. What's a simple way to avoid this problem?
State your key instructions and context once, clearly, near the start of a conversation.
Three things make the biggest difference: being specific (exact details, not vague requests), setting limits (how long, what tone, what to avoid), and giving an example (showing what you want, not just describing it).
Like this"I want a car" gets you nothing useful. "I want a 5-door estate, diesel, under £15,000, in blue" gets you an actual shortlist.
A vague request forces the AI to guess - and it will guess the most generic, average version, because that's the safest guess it can make.
Weak: "write me something about coffee"
Better: "Write a 90-word Instagram caption for a specialty coffee shop. Warm and a little playful, no emoji, end with one clear invitation to visit."
Type a short, vague version of your Stage 0 task, like:
Write me a quote for a customer.
Read the reply. Then, in the same chat:
Now make it 60 words, friendly but professional, and include a price of £250 and a start date of next Monday.
The second reply should feel far closer to something you'd actually send.
1. Name the three things that most improve a request.
Being specific, setting limits, and giving an example.
2. Why do vague requests get generic answers?
Because the AI fills in the gaps with the safest, most average-sounding guess.
Don't expect the first reply to be perfect. Treat it like a conversation, not a single request into a machine.
Like thisLike a test drive: you don't reject the whole car because the seat's in the wrong position - you adjust the seat.
If something's close but not quite right, ask a short follow-up to fix just that part, instead of throwing the whole thing away. It's faster, and each follow-up gives the AI more of a sense of exactly what you want.
If a reply is too formal, you don't need to rewrite your whole request - just say "make this sound more casual, keep it the same length."
Write a short email asking a customer to pay an overdue invoice.
Once you get a reply, don't start over. Type:
Make it firmer, but still polite.
Then:
Shorten it to 3 sentences.
Notice you never had to repeat the original request.
1. What should you do if the first answer isn't quite right?
Ask a short, specific follow-up question rather than starting over.
2. Why is this usually faster than starting again?
The AI already has the earlier context, so a small follow-up gets a small fix.
If you ask the AI to explain its thinking step by step before giving you a final answer, you often get a more accurate result - especially for numbers, logic, or several factors at once.
Like thisLike asking a mechanic to show you the fault before you agree to the bill.
Instead of "what should I charge for this job," try "walk me through your thinking step by step for pricing this job, then give me your final number." Now you can check if it miscounted your hours before believing the total.
There's a second habit: tell it to ask you a question first if anything is unclear, instead of guessing and running with it. Left alone, it will usually fill any gap with its best guess and answer anyway. That feels fast, but often means it solved the wrong version of your problem.
Before you answer, if anything here is unclear or you're missing information you'd need to get this right, ask me first instead of guessing.
Think of a decision from your work with more than one factor (pricing a job, choosing between two suppliers). Try:
I need to price a 3-hour landscaping job. My hourly rate is £25 and materials cost £40. Walk me through your thinking step by step, then give me a final price at the end.
Then try a request that's missing a detail, with the "ask me first" line attached:
Write a reply to a customer complaint. Before you answer, if anything here is unclear or you're missing information you'd need to get this right, ask me first instead of guessing.
It should ask what the complaint was about, instead of inventing one.
1. What does the AI usually do when your request is missing a detail, if you don't tell it to ask?
It fills the gap with its best guess and answers anyway, rather than stopping to check.
2. Why is asking it to check with you first often worth the extra step?
It costs a little time but avoids getting a confident answer to the wrong version of your question.
Telling the AI to "act as" something specific - a strict editor, a sceptical customer, a friendly teacher - changes how it responds. It's trying to match the way that kind of person would talk.
Like thisLike handing the same car to a boy-racer, a nervous first-time driver, and a professional reviewer - same car, each one notices completely different things.
Useful for tone or a different point of view. It does not give the AI secret extra knowledge. Style, not qualifications.
Ask it to "act as a sceptical customer" reading your website, and it will point out doubts a plain "helpful assistant" reply never would.
Copy a paragraph from your own website or an email you've sent. Paste it with:
Act as a generous, supportive reader. Here's my text: [paste your text]. What do you like about it?
Then, in the same chat:
Now act as a sceptical stranger reading this for the first time. What would put you off?
The second one usually catches things the first misses entirely.
1. What does giving the AI a role actually change?
Its tone and point of view - the style of the answer, not the underlying facts it knows.
2. True or false: giving it a role unlocks new information it didn't have before.
False - it changes how existing knowledge is presented, not what it knows.
If you give the AI one or two real examples of the style or format you want, it will copy that pattern far more reliably than if you just describe it in words.
Like thisThe difference between telling a garage "paint it a sort of deep reddish colour" and handing them the exact paint swatch.
This is the quickest fix when a reply is "close, but not quite the right shape."
Instead of "write product titles in my style," paste two of your own real product titles as examples, then ask for five more that match.
Find one sentence or short paragraph you've written for your business that you're happy with. Paste it with:
Here's an example of the style I want: [paste your example]. Write 3 more in exactly this style, about [your topic].
1. What kind of problem does giving an example solve best?
When a reply is close, but in the wrong style or format.
2. Why does an example often work better than a description?
It's much easier to copy a real pattern than to guess one from a written explanation.
Most chat AI apps let you save standing instructions that apply to every new conversation - sometimes called "custom instructions" or "preferences."
Like thisLike saving your seat position and radio presets in a car, so you don't reset them every time you get in.
This is still just chat. Nothing new is being connected or automated. It saves you typing the same background every time: your tone, your business, your formatting rules.
Saving "Always write in UK English, keep answers under 150 words unless I ask for more, and don't use exclamation marks" once, rather than typing it in every fresh chat.
In your AI app, find settings (usually a gear icon, or your account name) and look for "Custom instructions," "Preferences," or "How would you like [the AI] to respond?" Type:
I run a small landscaping business. Always reply in a friendly, plain-English tone. Keep answers under 100 words unless I ask for more.
Save it, then start a brand new chat and ask an unrelated question - notice the tone still matches, without you asking again.
1. What kind of thing belongs in a saved preference?
Anything you'd otherwise type at the start of every conversation - your tone, your context, your formatting rules.
2. Does saving preferences let the AI take actions on your behalf, like sending emails?
No - it only shapes how it replies to you in chat. Taking actions is a different, further step (often called an "agent"), which this course doesn't cover.
This is where Stage 0 to Stage 9 become one repeatable process for your real task - still entirely in chat, nothing new to install. The goal was never to memorise ten separate tricks.
Like thisThink back to Stage 0: you named the job you needed a car for. Everything since has been picking the right model, spec, and options for that exact job. This stage is driving off the forecourt with the whole thing built.
A clear task, the right assistant, a specific request, a real example, a check on the reasoning, a different point of view, a quick refinement - saved so you can reuse it every time this task comes up.
A landscaper's finished process: saved preferences set the tone and business details once (Stage 9) → a request with clear details for the quote (Stage 4) → two real past quotes as examples (Stage 8) → an "act as a sceptical customer" check (Stage 7) → two short follow-ups to polish it (Stage 5).
Open your Stage 0 note. Using the landscaper example as a guide, write your own single message that combines at least three ideas from this course - for instance your saved preferences (Stage 9) plus a specific request (Stage 4) plus one real example (Stage 8). Type it into a chat and see the result. Save this message in your notes - it's now a template you can reuse every time this task comes up.
1. What was the real goal of this whole course?
Not ten separate tricks - one repeatable process, built from combining them, for a task you actually do.
2. If you find yourself wanting the AI to check your emails or book something by itself, what does that mean?
You've reached the edge of what plain chat can do. That's a different area - often called "agents" - and it's a step beyond this course.
That's the full course. Enjoyed it, or have any feedback? Let me know.