If you read the first article in this series, you met the Librarian. Here is a quick reminder before we go further.
A quick recap: the Librarian behind the wall
Imagine a vast library staffed by one extraordinary Librarian. They have read every book, article, web page, and document that has ever been written — in every language. Their knowledge is genuinely remarkable.
The catch is that they work behind a solid wall. There is no window, no phone, no door. There is only a narrow mail slot cut into the wall at chest height. You can slip a note through the slot. A moment later, a note slides back out.
That is the whole system. One note in. One note out.
The Librarian's one job is to finish whatever is written on your note. They are, at the deepest level, the world's most sophisticated word-guesser — they predict, word by word, what the most useful and coherent continuation of your note would be.
And the instant the reply slides back, they forget you completely. No memory of any past note. Every exchange starts from a blank slate.
This is what the tech world calls a large language model, or LLM. The note you slip in is what everyone calls a prompt. That single word — "prompt" — is the subject of this article.
Why the note is everything
Here is the critical thing that trips up almost everyone who first uses an AI chatbot.
The Librarian cannot ask you to clarify. They cannot pop their head through the slot and say "sorry, did you mean X or Y?" They cannot send a follow-up note an hour later to check whether their answer was useful. They have no way to probe for what you actually needed.
They can only finish what is written on your note.
This is not a bug that will be fixed in the next update. It is the nature of the system. One note in, one note out. The entire skill of getting good results from AI rests on this single reality: if the note is vague, the reply will be vague. If the note is clear, the reply has a chance of being genuinely useful.
Think of it like ordering at a busy cafe. If you say "I'd like something warm," you might get tea, or soup, or a toasted sandwich — the server will make a reasonable guess. But if you say "a medium oat-milk latte, please, with an extra shot," you get exactly what you wanted. The cafe cannot read your mind. Neither can the Librarian.
What makes a good note: three things
Writing a good prompt does not require any special technical skill. It just requires thinking a little more carefully before you slip the note through the slot. Three things make the biggest difference.
First: be specific about the topic. "Tell me about the Romans" could produce ten thousand different notes in reply — a timeline, a recipe, a battle, a political analysis. "Explain how the Roman army was organised, in a way a twelve-year-old can follow" narrows it to something useful.
Second: give an example of what you want. The Librarian is brilliant at continuing a pattern. If you start the note with an example of the style or format you are after, they will pick up the thread. "Here is a sentence I like: 'The Roman legionary carried a short sword, a heavy shield, and enough food for three days.' Write me three more sentences in the same style about Roman soldiers" is far more powerful than a vague request.
Third: say what shape the answer should take. A paragraph? A bullet list? A short script? Three options, each with pros and cons? The Librarian is happy to give you any of these — but only if you ask. Without guidance, they will produce whatever felt most natural as a continuation of your words.
Before and after: two notes, two very different replies
Here is the same question written two ways. One vague, one clear. Read them and see which you would rather hand to someone who cannot ask a single follow-up question.
Weak note: "Help me with my history essay."
What the Librarian has to work with: the topic, the year group, the question, the length, the level of detail, how much has already been written, what kind of help is needed — none of this is on the note. The reply will be a generic guess, probably some broad advice that fits nobody in particular.
Clear note: "I am a Year 8 student writing a 400-word essay on why Henry VIII broke with the Catholic Church. I have already written my introduction. I am stuck on how to explain the political reasons clearly. Can you give me two or three points I could use, each in one short sentence?"
What the Librarian has to work with: year group, word count, essay topic, what has been done, exactly where the block is, the format of the answer. The reply is almost certain to be directly useful.
You do not need to write an essay every time you ask a question. But spending twenty extra seconds making your note more specific will save you far more time reading through an answer that missed the point.
How a "conversation" is actually faked
You may have noticed that when you use an AI chat tool, it seems to remember what you said a moment ago. You can ask a follow-up question and the AI appears to understand the context.
So what happened to "no memory"?
The trick is clever but simple. The chat tool secretly rewrites your note each time. When you send a second message, the tool does not just send that message through the slot. It writes out the entire conversation so far — your first message, the Librarian's reply, your second message — and sends the whole thing as one long note.
The Librarian receives something like this:
You said: "Explain why the Romans built roads."
I said: "The Romans built roads primarily for military movement — to get legions from one part of the empire to another quickly. They also served trade and communication."
You said: "Which road in Britain was the most famous?"
The Librarian reads this entire script and continues from the last line. They are not "remembering" the conversation — they are reading a written record of it, pasted into a fresh note, every single time you send a message.
This is elegant and remarkably effective. It also explains some puzzling things: why very long conversations sometimes start feeling confused (the note is getting enormous and important bits get buried), and why each conversation starts totally blank (there is no old note to paste from).
The third article in this series — How AI Fakes Memory — goes into this in much more depth. For now, the key point is that even the appearance of memory is just a note, cleverly written.
The whole skill in one sentence
Everything clever you can do with AI — every trick, every technique, every "prompt engineering" tip — is, at heart, just one thing: writing a better note.
Be specific. Give an example. Say what shape you want the answer in. And remember that the Librarian is brilliant, patient, and eager to help — but they cannot ask you what you meant. That part is on you.
This is part 2 of the AI in Plain English series for UK parents and KS3 students. Next up: How AI Fakes Memory — where we open the note to show what is really written on it when AI "remembers" your conversation.