Beginner · 07
Talking to AI: prompting well
A chatbot is only as good as what you ask it. Five habits that turn vague answers into great ones.
6 min read · Tap underlined words for a definition
Swipe up to begin
A prompt is everything you give a before it answers. The model has never met you, can't see your screen and doesn't know what you already tried. Most bad answers come from that leave it guessing.
1. Give context
Say who it's for and why. "Explain inflation" gets a textbook paragraph. "Explain inflation to my 12-year-old who just asked why sweets cost more than last year" gets something useful.
2. Show an example
If you want a particular style or format, paste one. Models are excellent at copying a pattern they can see, and poor at guessing one they can't.
This trick has a name: few-shot prompting. One or two examples often beat a paragraph of instructions.
3. Say what good looks like
Length, tone, format, what to leave out. "Three bullet points, plain English, no jargon, under 80 words" is a complete spec in one line.
Which prompt will most likely get the most useful answer?
4. Ask it to think first
For maths, logic or tricky decisions, ask the model to work through the problem step by step before answering. Writing out the steps gives it room to catch its own mistakes.
5. Iterate, don't restart
Treat the first answer as a draft. "Shorter." "Less formal." "Why did you suggest that?" The conversation is the tool.
And always check facts that matter. A well-prompted model is still a model: it can sound certain and be wrong. That's called a hallucination.
What is few-shot prompting?
Before you drift off
- Context, examples and a clear picture of a good answer do most of the work.
- Ask for step-by-step thinking on hard problems.
- Iterate on the draft, and verify facts that matter.
Terms in this lesson
Go deeper, free and open
- Prompt Engineering Guide · DAIR.AI · Free, MIT licensed
- Learn Prompting · Learn Prompting · Free beginner course
That's the lesson
Talking to AI: prompting well · 6 min
Spotted a mistake? Suggest an edit
Previous: Meet the chatbots: LLMs