How AI Chatbots Work: A Simple Explanation

The core idea: predicting the next word

An AI chatbot is built on a Large Language Model (LLM) — a program trained on enormous amounts of text. Its one job sounds simple: given the words so far, what word most likely comes next? It does this one word at a time, thousands of times, until a full answer appears. Because it learned from so much text, those predictions are often surprisingly smart.

Step by step: your message’s journey

• 1. Tokenising: your text is chopped into small pieces called tokens
• 2. Understanding context: the model looks at all the tokens together — your question, and the conversation before it
• 3. Predicting: it calculates probabilities for the next token and picks a likely one
• 4. Repeating: the new token joins the text, and the process repeats until the answer is complete

Where did it “learn” all this?

Before release, the model was trained on books, websites, code and more, using very powerful computers. Training adjusts billions of internal numbers until predictions become accurate. After training, extra steps teach it to be helpful and safe in conversation.

Why chatbots sometimes make things up

The model predicts likely-sounding text — it does not look up facts in a database. So it can invent a date, a quote or a reference that sounds perfect but is false. This is called hallucination, and it is the main reason to verify important answers.

How to get better answers

Give context, ask clearly, and request a format: “Explain photosynthesis in 5 bullet points for a 10th-grade student.” Better questions genuinely produce better answers — this skill is called prompt writing, and it is worth learning.

Go deeper — context window and temperature

Two ideas explain most chatbot behaviour. The context window is how much of the conversation the model can “see” at once — very long chats can push early details out, which is why it may forget them. Temperature controls randomness: low settings give careful, repeatable answers; high settings give creative, surprising ones. You do not set these directly in most apps, but they explain the results you get.

Practice task (30 minutes)

Ask the same question three ways: (1) one short line, (2) with your level and goal (“Explain like I am in semester 3”), (3) with a format request (“…in a table with examples”). Compare the answers and note what changed. This is prompt writing, practised properly.

Common mistakes to avoid

• Treating the chatbot as a search engine with guaranteed facts
• Asking one giant question instead of step-by-step questions
• Sharing personal, financial or password information in chats