Prompt Engineering cookbook

You ask for a short table and get three paragraphs back, then a bulleted list next time you ask the exact same thing.

The chatbot handed you a quote and a source that turned out not to exist.

Every answer comes back shaped differently, so you can never paste it straight into your report.

The reply reads like a bland encyclopedia entry when what you needed was a blunt opinion.

I told the chatbot never to mention a competitor, and it worked the name into a reply anyway.

The chatbot solved a word problem confidently and got the number wrong.

Why does one short question turn into five paragraphs of hedging?

Your prompt keeps producing bad output and you cannot tell which part of it is broken.

A snippet you found online does something you cannot explain, and your fix depends on it.

Flip the order of my few-shot examples, and the exact same input gets a different answer.

Asked a question about a topic I know well, and the chatbot's answer confidently mixed up two related rules.

You have rewritten the same request four times and it still is not landing right.

Asked the same hard reasoning question twice, got two different answers, and had no way to know which one to trust.

A prompt that worked well on one chatbot falls oddly flat on a different one.