If you think of an AI like a giant, digital brain, Prompting is simply the art of asking the right questions to get the best answers. Again let’s think of the AI as the brilliant but literal intern. If you give a vague instruction, you’ll get a vague result. If you give clear, clever instructions, you’ll get magic. Here is how you can start to “train” your AI intern using a few simple techniques.
1. The Basics: Setting the Stage
Most people start here. These techniques are about being clear and direct. Shot Prompting
- The Direct Ask (Zero-Shot): Just ask the question. It’s like asking a librarian for a fact.
- Example: “What is the capital of France?”
- Give a Few Examples (Few-Shot): If you want the AI to answer in a specific way, show it a few “flashcards” first.
- Example: “Happy means Positive. Sad means Negative. Frustrated means…” (The AI will say “Negative”).
- The Costume Party (Role-Playing): Tell the AI who it should pretend to be. This changes its “personality” and the words it uses.
- Example: “Act as a friendly grandmother and explain how to bake a cake,” vs. “Act as a drill sergeant and explain how to bake a cake.”
2. Sharpening the Results
To move from “okay” answers to “great” ones, you need to add boundaries.
- Specific Instructions: Don’t just say “Write about dogs.” Say, “Write three funny paragraphs about Golden Retrievers, focusing on how much they love tennis balls.”
- Contextual Framing: Tell the AI who the audience is.
- Example: “Explain how the internet works, but pretend I am five years old.”
The more clear and specific you are in the instructions the better the end results with AI, just like in life. AI gives us the chance to reduce the manual tedious portions of things to mere minutes or seconds. But you will only get quality out, if you put quality in.
3. Advanced Moves: Making the AI “Think”
Sometimes the AI moves too fast and makes mistakes. These techniques will slow it down and make it more logical.
- Step-by-Step Thinking (Chain-of-Thought): If you have a hard math problem or a complex logic puzzle, tell the AI: “Think through this step-by-step.” This forces it to “show its work,” which usually prevents it from making silly errors.
- The Brainstorm (Tree of Thoughts): Ask the AI to come up with three different ideas, argue the pros and cons of each, and then pick the winner. It’s like having a board meeting inside the AI’s head. I have done this in the Information Technology space when I need to get up to speed on a new technology, and maybe 2 or 3 vendors in the particular space.
- The Self-Interview (Self-Ask): This tells the AI to ask itself the “missing questions” first.
- Example: If you ask about the weather in a specific zip code, the AI will first ask itself, “Where is this zip code?” and then “What is the weather there?”
- The Research Assistant (RAG): This is when you give the AI a specific document (like a PDF or a news article) and say, “Only use the information in this paper to answer me.” It stops the AI from guessing.
Summary Table: Which Tool to Use?
| If you want… | Use this technique… |
| A specific tone | Role-Playing (The Costume Party) |
| A specific format | Few-Shot (Give Examples) |
| Better logic/math | Chain-of-Thought (Step-by-Step) |
| A creative choice | Tree of Thoughts (The Brainstorm) |
One last separate “pro tip” if you sill is this:
You do not need to craft or create your entire prompt in the window given to you by Chat GPT, Claude, or Gemini.
Go ahead and dust out that old edition of Microsoft Word and create your prompts in a full windowed document that you can read from start to finish without scrolling.
