Iterate, then put it to work

No one writes the perfect prompt on the first try — not you, not the teams behind your favorite AI products. What separates them isn't better first drafts; it's treating prompts the way engineers treat code: draft, test, fix, repeat.

The loop is simple:

• Test on real inputs — not just the friendly example you wrote it for. The weird ticket, the empty field, the sarcastic customer.
• Change one thing at a time. Rewrite everything at once and you'll never know which change helped.
• Keep your failures. Every input that broke the prompt joins your private test set. A prompt that survives ten ugly real cases beats one that aced a demo.
• Say what to fix, not just "be better." When output disappoints, name the gap: too long? wrong reader? missed the edge case? Each complaint maps to a block from Chapter 5.

Do this a few times and prompting stops feeling like luck. Then the skill starts paying rent everywhere — scroll through where.

You now know how to talk to an AI

The whole course in six lines:

• The model completes your text — the prompt is the steering wheel, not a wish.
• Be specific: every detail you give is a guess it doesn't make.
• Show examples when the pattern is easier to demonstrate than describe.
• Give it room to think on anything multi-step.
• Structure serious prompts — role, context, task, examples, format, guardrails.
• Iterate. Prompts are drafts until they've met real inputs.

None of it is magic — it's clear communication with a machine that predicts text. If you want to go one level deeper into that machine, the How ChatGPT Actually Works course explains the predictor you've just learned to steer — and the other courses go deeper still.