Prompt

What is a prompt?

A prompt is the input a person gives a generative AI — such as a large language model like ChatGPT — to make it carry out a specific task. It is usually natural-language text, but it can also include examples, data or an image. A prompt can be as short as a single question or as detailed as a multi-part instruction specifying role, context and output format.

Why does the prompt matter so much?

Generative models produce their answer based on the input. A precise, well-structured prompt therefore strongly shapes the quality of the result: it gives the model context and makes clear exactly what is expected. Vague prompts, by contrast, often yield vague or off-target answers. The deliberate design of inputs is called prompt engineering. One advantage: unlike retraining the model, prompting does not change the model's parameters — you only adjust the input.

What makes a good prompt?

A few principles help: a clear task rather than a vague request, context (background, audience, purpose), the desired format (length, structure, tone) and, where useful, examples of the expected output. For studying, role and task instructions work well, such as "Explain the Pythagorean theorem as if I were in 8th grade, with an everyday example" or "Ask me five comprehension questions about this text and grade my answers."

A few terms help newcomers: giving no examples is called zero-shot prompting, while providing a few examples is few-shot prompting. Many applications also separate a system prompt (a fixed role and rules) from the actual user prompt.

One caveat remains: even a good prompt does not guarantee a correct answer, because language models can fabricate facts. Results should therefore be checked. Our guide on studying with ChatGPT shows how to use prompts deliberately for learning.