Why this lesson exists. Everything else in this course depends on one idea, so it comes first. Get it and the rest follows; miss it and AI will keep disappointing you in ways that look random.
By the end of this lesson you will be able to explain why the same AI tool gives a generic answer one minute and a superb one the next, because you will have produced both yourself and seen what made the difference.
Open whichever AI tool you have. Type:
Write a short summary of our company for a new starter.
Read what comes back. It will be fluent, well-structured and almost entirely generic: "a dynamic company committed to delivering value to its customers". It could describe anyone. It describes no one.
Now paste in your company's about page, a product description, or a recent pitch, and ask exactly the same question again.
This time you get something you could actually give a new starter. Names are right. Products are right. The tone is closer to yours. Nothing about the tool changed between the two attempts. The only thing that changed is what it could see.
An AI model of the kind you use at work (Claude, ChatGPT, Gemini, Mistral's Le Chat, Copilot) is a piece of software that has read an enormous amount of public text and learned the patterns in it. When you type something, it produces the most useful continuation it can, based on those patterns plus whatever is in front of it right now.
That second part is the key. The model has read the public library. It has never been inside your office. It does not know your prices, your clients, what you decided last month or how you like to sound. So when you ask it something that depends on those things, it does what a bright new colleague with no context would do: it produces something plausible. People call this "hallucination". It is more useful to think of it as a confident guess to fill a gap you left.
The technical word for everything the model can see when it answers is its context: your message, anything you paste or attach, any saved instructions that are loaded, and (later in this course) anything it can reach through a connection to your systems. The quality of the answer tracks the quality of the context almost exactly, because the model has nothing else to work from.
This is why the rest of the course is shaped the way it is. Lessons 2 to 4 are about putting the right things in your message. Lessons 5 to 7 are about building a store of company knowledge the model can draw on every time. Lesson 8 is about letting it look at your live systems. Lessons 11 onward are about doing all of that without you at the keyboard. It is all one idea, applied in bigger and bigger ways: get the right information in front of the model.
Because the model works from what it can see, it will repeat whatever you give it with total confidence, right or wrong, current or stale. Feed it last year's price list and it will quote last year's prices in a customer email without hesitation. That cuts both ways, and Lesson 7 is entirely about the second edge of it.
