Cortexa AI Glossary · Asking well
What is a prompt, and what is prompt engineering?
From Cortexa Learn, by Cortexa Consulting. Last checked .
What you type shapes what you get. Asking well is most of the skill.
A card for anyone
A coworker is retiring, and the card is going around. You open a chatbot and type it the way you'd type a search: retirement card message. Back comes something warm and smooth. "Thank you for your years of dedication. Enjoy this exciting new chapter." Nothing in it is wrong. It just isn't about the person you've sat next to for eleven years.
What a prompt is
What you typed was a prompt. A prompt is whatever you give an artificial intelligence (AI) tool to work from: the words you type, plus any file, photo, or example you attach. Your follow-up messages count too. Prompt engineering is the name for getting good at writing them. It sounds technical. Mostly, it's the skill of asking clearly.1
Why vague gets generic
A chatbot writes by predicting likely words. When your request leaves gaps, it fills them with the most likely guess. And the most likely retirement message is a lot like every retirement message it has ever seen. So a short prompt tends to get a safe, middle-of-the-road answer. Ask for a birthday message with no details, and you'll get one that fits any birthday, for anyone, from anyone. Nothing went wrong there. The chatbot did its best with what it had.
Four things to say
A clear request usually covers four things.
- The job: what you want made, like a short message for a card.
- The reader: who it's for, like a teammate retiring after eleven years who loves a bad pun.
- The materials: anything it should use, like a story from the office or a detail you want in.
- What good looks like: how long, what tone, and anything to leave out.
None of that is technical. It's what you'd tell a friend who offered to help you write it.12
Show one
An example often says more than a description. If you want the message short and a little funny, paste in one you liked from another card. Anthropic calls examples one of the most reliable ways to steer the tone, format, and structure of an answer. One good example can save a lot of explaining. It works the other way too: paste in the generic message you got, and say what's wrong with it.2
Keep talking
A prompt doesn't have to be perfect the first time. If the answer is close, say what to change. Make it shorter. Make it warmer. Mention the fishing trip. The chatbot can see the earlier messages in the conversation, so you can steer it instead of starting over. Each round adds a little more of what you know and it doesn't. A round or so of that often gets you there, and it's quicker than writing the perfect prompt up front.
No magic words
You may see lists of secret phrases that promise better answers. You don't need them. Clear writing does most of the work. Anthropic's advice is simpler: picture handing your prompt to a smart colleague who knows nothing about the task. If they'd be confused, the chatbot probably would be too. The next time you sign a card, try telling the chatbot who the person is before you ask for the words.2
Works cited
- IBM, "What is prompt engineering?" (checked )
- Anthropic, "Prompt engineering overview" (Claude Docs) (checked )