Prompt Engineering: How to Get Great Answers from AI

Here's something you'll notice fast: two people can ask ChatGPT the "same" question and get wildly different results — one gets a vague paragraph, the other gets exactly what they needed. The difference isn't the AI; it's the prompt. In Lesson 20 you learned an LLM predicts the next token based on your input, so your input steers everything. Prompt engineering is the skill of asking well — and it's arguably the highest-return AI skill you can learn today. No coding required.

What a prompt really is

A prompt is just your instruction to the model. Because the LLM continues from what you give it, a clear, detailed prompt narrows it toward the answer you want, while a vague one leaves it guessing. You're not "programming" — you're communicating, the way you would to a very fast, very well-read assistant who takes you completely literally.

The anatomy of a great prompt

Most strong prompts contain four ingredients. You don't always need all four, but keeping them in mind instantly upgrades your asking.

Technique 1: Be specific

Vague in, vague out. Tell the model exactly what you want — topic, length, audience, angle.

Technique 2: Give context

The model knows nothing about your situation unless you tell it. A sentence of background changes the whole answer.

Technique 3: Assign a role

Telling the model who to be shifts its tone, depth and vocabulary. "Act as a…" is one of the most powerful two-word upgrades you can make.

Technique 4: Show examples (few-shot)

If you want a specific style or pattern, show one or two examples rather than describing it. This is called few-shot prompting, and it's remarkably effective.

Technique 5: Ask for step-by-step reasoning

For anything involving logic, math or planning, adding "think step by step" pushes the model to work through the problem instead of blurting a guess — often turning a wrong answer into a right one. (This is the everyday version of chain-of-thought prompting.)

Technique 6: Specify the output format

Want a table? A JSON object? A bullet list of exactly five items? Ask for it explicitly and you'll get something you can use directly.

Iterate — it's a conversation, not a slot machine

Your first prompt rarely needs to be perfect. Treat the reply as a draft and refine: "shorter," "more formal," "add a real-world example," "now rewrite it for a 10-year-old." Because the model remembers the conversation (its context window from Lesson 20), each nudge builds on the last. Great results usually come from the third or fourth message, not the first.

Common mistakes to avoid

Putting it all together

Here's a prompt that stacks role, task, context and format into one clear request — the kind that reliably gets a genuinely useful answer:

The golden rule: the AI can't read your mind — it can only read your prompt. The clearer your ask, the better your answer.

Prompt engineering is the skill that made AI genuinely useful in my day rather than a novelty — once I started writing prompts with a role, clear task and a format I actually wanted, the quality jump was night and day, and I stopped blaming the tool for lazy questions. I'd suggest saving 3–4 prompts that work well for you as reusable templates; that small habit compounds fast.

Key takeaways

• Prompt engineering = communicating clearly with an AI; it's the highest-ROI, no-code AI skill.
• Strong prompts often combine Role + Task + Context + Format.
• Be specific, give context, assign a role, and show examples (few-shot) for a target style.
• Add "think step by step" for logic problems, and specify the exact output format you want.
• Iterate — refine across messages, and always verify important facts.

Continue the series: ← Lesson 20: Large Language Models (LLMs)  ·  Next: Lesson 22 — RAG: Giving AI Your Own Data (and Fewer Hallucinations) →