What is Prompting?
First, what is prompting? Prompting, or prompt engineering, is the practice of designing clear, structured inputs to guide AI in generating accurate and useful responses. Instead of asking vague or open-ended questions, effective prompting ensures AI understands context, tone, and format, ultimately leading to better results.
Getting useful responses from AI isn’t just about asking a question—it’s about asking the right question. If you’ve ever found AI-generated responses to be too vague or not quite what you were looking for, refining your prompts is the key. A simple framework I learned from Google Prompting Essentials is the TCREI framework: Task, Context, References, Evaluate, and Iterate.
What is the TCREI Framework?
The TCREI framework is a structured approach to crafting effective AI prompts. It ensures that the AI understands your request by breaking it down into five essential components:
- Task: Define what you want AI to do, including specifying a persona and desired output format.
- Context: Provide relevant details that help AI generate more accurate and meaningful responses.
- References: Give examples or benchmarks to guide AI’s response style and structure.
- Evaluate: Review AI’s output to determine if it meets your needs.
- Iterate: Refine and tweak your prompt to improve results.
However, I believe this framework can be simplified further because the last two items, Evaluate and Iterate, are more intuitive and tend to happen naturally. Including them may overcrowd the framework, so they can potentially be dropped.
Now, let’s take a deeper dive into the renewed framework of TCR – Task, Context, and References.
The TCR Framework
The Prompt Item (TCR) | Explanation | Considerations | Example |
|---|---|---|---|
Task | Define what you want AI to do. Specify a persona and a format. | Who is the AI emulating? What output format do you need? | “As a nutritionist specializing in plant-based diets, generate a weekly meal plan in a table format, including breakfast, lunch, and dinner.” |
Context | Provide necessary details to refine the output. | What specific details would improve AI’s understanding? | “Provide five easy-to-make vegetarian meal ideas for a college student with a limited budget and access to basic kitchen appliances.” |
References | Include examples to guide AI’s response. | How many references are ideal? Do they shape the tone, style, or content? Shots mean “references” or “examples.” – few-shot prompting (multiple examples) – single-shot prompting (one example) – zero-shot prompting (no examples) | “Generate five meal plans based on these two references:…” |
- Now, you will naturally evaluate whether the AI’s response meets your needs.
- If not, you will need to iterate or tweak your prompts to enhance results. You may need to simplify or add more detail.
Final Takeaways
- Start simple, then gradually add complexity as needed.
- Keep experimenting—great prompting is a skill that improves with practice!
On a final note, Google recommends the TCREI framework (Task, Context, References, Evaluate, and Iterate), and to simplify recall, they introduce the phrase “Thoughtfully Create Really Excellent Inputs.” Although my revised TCR (Task, Context, References) framework is simpler enough to remember, I will go ahead and recommend “Truffle, Chocolate, Rum“. That decadent dessert is hard for me to forget!
Ayse Ozturk
Sources:
- Introduction to prompting, https://cloud.google.com/vertex-ai/generative-ai/docs/learn/prompts/introduction-prompt-design
- Google Prompting Essentials, https://grow.google/prompting-essentials/
Leave a Reply to Simple Tips for Using ChatGPT – Dr. Ayse OzturkCancel reply