Large Language Models (LLMs) such as ChatGPT, Claude, and Gemini generate text by predicting the next token from patterns learned in massive amounts of training text, not by looking up facts. This lesson covers tokens, context windows, and the six-part prompt structure (role, task, context, rules, examples, format) for writing prompts that get useful, specific answers.
What you will be able to do
- Write prompts using the six-part structure: role, task, context, rules, examples, format
- Estimate roughly how many tokens a prompt uses (about 750 tokens per 1,000 words) and keep it within the context window
- Explain why LLMs hallucinate and apply the safety tips (treat the model as an assistant, set boundaries, ask for sources) to reduce risk
- Apply the three-step formula, name the role, state the task with a verb, set rules and format, to turn a vague prompt into a clear one
- Run a prompt against the good-prompt checklist before sending it
Before you start
- Access to an LLM chat tool such as ChatGPT, Claude, or Gemini
- Basic familiarity with typing a message into an AI chat interface
- No coding or technical background needed
Reference
| Prompt part | What it does | Example from the lesson |
|---|---|---|
| Role | Sets who the model should pretend to be | "Act as a science teacher for grade five" |
| Task | States what should be done, using a verb | Summarize, rewrite, compare, or plan |
| Context | Gives facts the model needs | Notes, data, or short passages |
| Rules | Sets boundaries on the answer | "Keep it under 150 words," "use simple words" |
| Examples | Shows the kind of output wanted (few-shot) | One or two sample inputs with expected answers |
| Format | Tells the model how to shape the answer | Paragraphs, bullet list, JSON, or table |
| Token rule of thumb | Estimates prompt/response size for cost and context limits | About 1,000 words equals 750 tokens |
Common errors and fixes
| What goes wrong | The fix |
|---|---|
| Prompt is vague and long | Cut fluff, state the task with a verb, add one or two rules, then stop |
| No audience is set | Say who the reader is: age, role, or skill level |
| No format is given | Ask for a structure like bullets, table, JSON, or clear sections |
| Too many goals in one ask | Split into small steps: plan, then draft, then polish |
| Model gives a wrong answer with confidence (hallucination) | Ask for sources when they matter, and stop to rethink the task if the answer seems wrong or unsafe |
Read the full walkthrough
The complete lesson, with screenshots and any downloads, is published on Substack as part of Prompt Engineering for AI Automation.
More in this section
- Lesson 2: How to control the large language models output?
- Lesson 3: Zero-Shot, One-Shot, and Few-Shot Prompting Explained
Continue the course
Browse all lessons in the Prompt Engineering for AI Automation course, or subscribe to the GenAI Unplugged newsletter to get new lessons in your inbox.
