Y9 Introduction to Prompt Engineering

On the previous page, you learned that a Large Language Model uses the words in a prompt as context and predicts what is likely to come next.

A prompt is the instruction, question or information you give an AI tool.

The information in a prompt influences what the AI generates. A clearer prompt usually produces a response that is more relevant and useful—but it does not guarantee that the response will be correct.

Task #1 – Revisit the Human versus AI response  

Teacher-led class activity 

In the previous lesson, your teacher entered this story starter into Gemini:

When the school caretaker opened the cupboard, they discovered something that definitely should not have been there…

Your teacher will reopen the Gemini chat from that lesson and show you:

• the original prompt

• the response Gemini generated

Take a moment to reread the prompt and response so you can remember what happened.

Discuss as a class

• What choices did Gemini make because we had not given it clear instructions?

• What did Gemini not know about the story we wanted?

• What parts of the response might we want to control?

• What could we add to the prompt to give Gemini clearer directions?

Task #2 – Improve the prompt

As a class, rewrite the prompt to give Gemini clearer directions.

Consider:

• What type of story should it be?

• Who is the story for?

• How long should it be?

• What should it include?

• How should it end?

• Is there anything it should avoid?

An improved prompt could be:

Continue this story as a funny mystery for Year 9 students:

“When the school caretaker opened the cupboard, they discovered something that definitely should not have been there…”

Write between 80 and 100 words. Include a strange sound and a harmless surprise. 

End with a cliff hanger. Do not include violence.

Your teacher will open a new Gemini chat and enter the improved prompt.

A new chat is used so Gemini responds only to the improved prompt and is not influenced by the earlier conversation.

Compare the two responses

• What changed in the new response?

• Which added instruction had the clearest effect?

• Which response was better suited to the audience and purpose?

• Did Gemini follow every instruction?

• Is there anything you would still change?

On the previous page, you learnt that a Large Language Model uses the words in a prompt as context to predict what is likely to come next.

A useful prompt gives Gemini enough relevant information to guide those predictions towards the type of response you need. Gemini does not automatically know your purpose, audience or preferred format. When these details are missing, it makes choices based on patterns learned during training. Adding clear, relevant instructions gives you more control over the direction of its response. 

Think about these four questions when writing a prompt:

You will not always need to include every possible detail. The aim is to give Gemini the information that is relevant to your task, audience and intended result. 

Now that you know the four parts of a useful prompt, see whether you can find them in an example.

Task #3 – Prompt Detective

Work with a partner

Read the prompt below:

Suggest three achievable ideas for a simple 2D Scratch game aimed at students aged 11–13. I am a Year 9 beginner and have four weeks to create the game. For each idea, include the game’s goal, player controls and main challenge. Use no more than 60 words per idea. Do not include online multiplayer.

Find the following parts:

• the task

• the context and audience

• the requested output

• the rules or limits

Be ready to share your answers with the class.

Task #4 – Fix a vague prompt 

Work with a partner

Choose one of the following prompts:

• Give me a game idea.

• Design a character.

• Help me study for a test.

• Write a story.

• Explain how a computer works.

• Tell me about space.

Task #5 – Share what you discovered

Your teacher will invite several pairs to share:

• their original prompt

• their final improved prompt

• the most noticeable change in Gemini’s response

Discuss as a class

• Which prompt changes had the strongest effect?

• Did adding more words always make a prompt better?

• What kinds of information were most useful to Gemini?

• Did Gemini always follow every instruction?

• Why do we still need to check an AI-generated response?