# Prompt Engineering Basics

> What a prompt is and what prompt engineering adds. The three jobs a good prompt does, and why one good answer proves very little.

**Course:** [Prompt Engineering](https://learn.benibix.com/learn/ai-working-with-ai)  
**Time:** 10 min

**After this lesson you can:**

- Say what a prompt is and how a model reads it
- Write a prompt with a task, facts and a shape
- Explain why a prompt is tested on many inputs

## What a prompt is, and what makes one good

A prompt is the text you send to an AI model to say what you want. Prompt engineering is writing that text, testing it and fixing it until the answers come out right. The model reads your prompt as tokens. In English, one token is about four letters, so 100 tokens is about 75 words.

The model behind a chat app is a large language model, or LLM. It writes by guessing what text should come next, given all the text above. So a prompt is not a search box. Every word you send shapes what comes back, the way the first line of a story shapes the rest.

A good prompt does three jobs. It says the task. It gives the facts the model cannot guess, such as who will read the answer. And it shows the shape of the answer: how long, in what format, in what tone. Most weak answers come from a prompt that skipped one of the three.

Be direct and be specific. Write about dogs leaves the model to guess the reader, the length and the point. Write 100 words for a 10-year-old on why dogs pant leaves almost nothing to guess. The second prompt is longer, and that is fine. Clear beats short.

Prompt engineering is a loop, not one try. You write a prompt, run it, read the answers and change one thing. Then you run it again. One good answer proves very little. A prompt that will be sent a thousand times has to be tested on many inputs, not one.

Tokens matter because they are what you pay for, and what fills the model's space. Every limit and every price is counted in tokens. A long prompt costs more each time it is sent. So the aim is not the longest prompt. It is the shortest one that still says all that is needed.

This course is about the words you send and what comes back. It stops at the prompt. Training a model on your own data, building a search over your own files and writing an agent from scratch are separate subjects. Coding with AI is in here, because there the prompt is still the main tool.

```text
Weak:
Write about dogs.

Strong:
Write 100 words for a 10-year-old on why dogs pant.
Use short sentences and one fun fact.
End with a question the child can answer.
```
The strong prompt names the reader, the length and the shape. The weak one leaves all three to chance.

![One prompt drawn as a single block in four parts: the task, the facts, a few examples and the question. An arrow runs from the block to the model, and on to the answer.](https://learn.benibix.com/diagrams/pr-prompt-parts.svg)

### Example: Lucy writes a lamp blurb

Lucy types describe this lamp and gets 230 words of praise. Her shop page has room for 40 words.

She adds the reader, the length and two facts: the lamp is 30 cm tall and runs on batteries. The next answer is 38 words and fits the page first time.

**Words in this lesson:** [Prompt (AI)](https://learn.benibix.com/term/ai/prompt), [Prompt Engineering](https://learn.benibix.com/term/ai/prompt-engineering), [Large Language Model (LLM)](https://learn.benibix.com/term/ai/large-language-model), [Token (AI)](https://learn.benibix.com/term/ai/token)

**Question:** What makes an AI prompt good?

**Answer:** It says the task, gives the facts and shows the shape of the answer. The model cannot guess what you left out. Naming the task, the facts it needs and the shape you want removes the guessing. Short or polite does not.

---

Source: https://learn.benibix.com/learn/ai-working-with-ai/prompt-engineering-basics
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-20
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
