How LLMs Read Your Words

You do not need to know how an engine works to drive well — but knowing a few basics makes you a much better driver. This lesson gives you the three non-technical facts about AI models that explain why some prompts work and others flop.

Fact 1: The AI reads in tokens, not ideas

Large language models (LLMs) — the technology behind Claude and ChatGPT — break your text into small chunks called tokens. A token is roughly a short word or a piece of a longer word. "The bakery is open" is about five tokens.

The model then predicts, token by token, what text should come next — based on patterns learned from enormous amounts of writing. It is astonishingly good at this, which is why it feels like understanding. The practical takeaway: the model responds to the actual words on the page, not to what you meant. If your prompt says "make it better," the model must guess what "better" means to you. If it says "make it shorter, friendlier, and mention free delivery," there is nothing to guess.

Fact 2: The context window is the AI's working memory

Everything in your conversation — your prompts, the AI's replies, any documents you paste in — sits in a space called the context window. Think of it as the model's desk. Whatever is on the desk, it can use. Whatever is not, effectively does not exist.

Two business-relevant consequences:

• The AI does not know your business unless you tell it. It has never seen your price list, your brand voice, or last week's customer complaint — until you paste them into the conversation.
• Very long conversations can drift. In a marathon chat session, early details can get less attention. If the AI seems to have forgotten your instructions, restate the key ones or start a fresh chat with a clean summary.

The context window is big — you can usually paste in whole documents, meeting transcripts, or several pages of notes. Use that. A consultant who pastes in her raw meeting notes gets a far better summary than one who writes "summarize my meeting" from memory.

Fact 3: Wording steers the output more than you would expect

Because the model predicts text from patterns, the style of your prompt shapes the style of the answer. Write casually and you tend to get casual answers. Use precise business vocabulary and you get more professional output. Ask "why is my marketing bad?" and you invite generic criticism; ask "list three specific reasons this email might get low clicks, then rewrite it" and you invite useful work.

A few wording habits that pay off immediately:

• Use numbers: "3 options," "under 150 words," "5 bullet points." Vague sizes produce vague output.
• Name the audience: "for first-time customers," "for my kitchen staff," "for a landlord."
• Say what to do, not only what to avoid. "Use short, plain sentences" beats "don't be wordy."
• Put important instructions at the start or end of a long prompt, not buried in the middle.

The mental model to keep

Picture a brilliant, endlessly patient temp worker with no memory of your business, who reads exactly what you hand them and starts writing immediately. Everything in this course — structure, examples, iteration, templates — is just an efficient way to hand that worker the right brief.

Try it now

Open your AI tool and ask: "What should I post on social media this week?" Note how generic the answer is. Now paste in a short description of your business, your typical customer, and one thing happening this week, then ask again. You just experienced the context window firsthand — the model did not get smarter, it just got the right things on its desk.