Anatomy of a prompt
About this app
Combine role, task, context, format and examples as stackable 3D blocks. Toggle components to improve a sample prompt without making AI calls.
What a good prompt is made of
A language model can only work with what the prompt contains. Five building blocks help you phrase a request completely. The role sets the perspective and tone of the answer. The task states the goal with a precise verb. The context supplies background, audience and constraints. The format specifies the output structure, such as a list, table or prose. Examples show concretely what kind of answer is wanted. Not every prompt needs all five, but every missing block leaves a decision to the model.
What you see
The five building blocks appear as stackable 3D blocks. Next to them is a composed sample prompt asking a maths teacher persona to write percentage word problems for year 7. Each block you switch on adds its sentence, and a “completeness” meter grows with it. Tapping a block explains what it is for. The app does not call a language model.
Things to try
Switch all blocks off and turn on only “task”: the prompt says what to do, but not for whom or in what form. Add context and format and read which questions are now answered. The completeness meter is a rule of thumb and is no substitute for testing with a real model.
Frequently asked questions
- How do I write a good prompt?
- Describe the task with a clear verb, give the necessary context such as audience and constraints, and specify the output format you want. A role and a short example help the model match tone and type of answer. After that, check the result and refine the prompt where needed.
- Why give an AI a role in the prompt?
- A role such as “You are an experienced maths teacher” steers perspective, vocabulary and depth of the answer. It does not replace a concrete task or context; without those, the answer stays generic despite the role.
- What is few-shot prompting?
- In few-shot prompting the prompt contains one or more examples of the desired input and output. The model follows the style, length and structure of those examples. For unusual formats this is often more effective than a long description.
- What can I do in this app?
- You switch the five blocks role, task, context, format and examples on individually or all at once and see the sample prompt grow and the completeness meter rise. Tapping a block shows its explanation. No AI requests are made.
Subject: Machine Learning
More from Machine Learning
- AdaBoost
- Data privacy traffic lights
- Decision Tree
- Ensemble: Bagging vs. Boosting
- Finding faces: masks versus learning from examples
- k-Means
- k-Nearest Neighbors (kNN)
- AI theatre
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps