Person holding phone with ChatGPT app open

A Guide to Prompting

This page is a free guide, showing you how to write effective prompts and utilise various prompting techniques.

Improving your prompting will help you to create more realistic, engaging and relevant learning experiences using generative AI tools, like ChatGPT.

Getting the most out of this guide

This comprehensive guide is designed to help anyone improve their prompting to get better results with generative AI.

The first two sections cover how to write effective prompts, but you shouldn't stop there. You won't always get the output you need first time, so section three shows you prompting techniques you can use to guide the AI to a better response. The final sections give you tips on how to prompt AI image generators and further resources to learn more.

Test out the advice for yourself and bookmark the page for future use.

Select the pink bookmarks directly below to jump to the section or scroll down to get started.

01.

How to craft effective prompts

02.

Other resources

How to craft effective prompts

Prompting AI is like giving instructions to a smart student or intern. So as learning professionals, we can tap into our know-how of giving clear and appropriate instructions.

Generative AI models pay attention to key terms in your prompt, predicting the best answer by drawing on the data from its network — as explained on the AI literacy page. So, how you write your prompt and what you include can greatly affect its output.

When writing prompts, it helps to be SIC:

Specific

state exactly what

you want

Informative

provide helpful context and relevant information

Clear

give instructions that are

easy to follow

The text and screenshot below is a simple example of a SIC prompt. It specifically tells the AI it's role and task. It then provides context before giving it a clear and specific instruction.

'You are an experienced secondary ICT teacher and AI expert. I need your help designing a simple 10 lesson unit about AI for grade 7 students.

The aim of the unit is to introduce students to artificial intelligence. The unit should include the basics of what AI is, ethical considerations relevant to their age and some useful, relevant and practical examples. The students will have heard of AI, but design a basic unit for beginners.

Write a unit overview titled 'Grade 7 AI Literacy', using sub-headings and bullet points.'

Your prompts might need to be longer for more complex tasks or if you include examples. You may sometimes want to give less information and see what the AI comes up with.

Think about the output you want and the information that will help the AI generate this.

What to include in your prompt

There’s no perfect prompt, but it's helpful to include some or all of the following, especially at the start of a project or for a more complicated task.

Role — the role or persona you want the AI to take (if you're not sure what this should be, ask it which expert(s) are most suited to the task).

Goal — the purpose of the project or task.

Context — useful information to increase the relevance and focus of the AI’s output.

Examples — show what you want the output to look like or include, e.g., tone of voice or structure.

Rules — specific criteria you want it to adhere to.

Instructions — be clear and specific about what it needs to do.

Before writing the prompt, it’s helpful to think about what you want the end output to be and how you will prompt the AI to get there. Typing longer prompts in a virtual notebook can help you organise your ideas.

Now you know how to write effective prompts, you can get more useful, relevant and insightful results from generative AI tools. Still, you may not always get the output you want. When this happens, you’ll need to tweak your instructions, give feedback or change tact — just like a skilled teacher!

In the next section, we’ll go through some useful techniques to help you master the art of prompting.

Prompting techniques

Your first prompt won’t always work as expected. But, don’t just accept a sub-par response. Use some of the techniques in this section to help you get better results.

Edit and regenerate

If you’re not happy with an answer, some AI tools, like ChatGPT, allow you to go back and edit your prompt. I did this for many of the prompts on this page!

Rather than just regenerate the answer, consider how you can edit your original prompt to get a better response.

In the following example, I wanted a hypothetical project I could use on this page. The output was OK, but I wanted more specific information and didn’t need the last section about Generative AI, which it must have misread from my prompt.

To get a better result, I edited my prompt by selecting the pencil icon at the top right of the prompt. I removed the text in my prompt about generative AI and gave more specific task instructions in the second and third paragraph.

Use delimiters

In long prompts, it can help to use symbols to organise your prompts and differentiate certain content, like examples. This helps to make sure that the AI reads your prompt correctly - and it’s easier for you to check and edit your prompt.

You can use any symbol, like < > or [ ], though it might be best to avoid commonly used symbols (like round brackets) which may appear elsewhere in your prompt.

In the prompt below, there’s a long example as well as content that the AI needs to analyse. I used delimiters to highlight these and so I could put the long text after my instructions. If you know how to use markdown, you can use markdown in your instructions.

Get it to respond ‘OK’

Do you get annoyed when the AI responds with a long answer when you don’t need it to? Or want to break down longer prompts to get better results?

Use the exact phrase: “Read the information and just respond with the words ‘OK’.” Then, it will read your prompt and respond ‘OK’.

This is a good way to feed the AI contextual information one prompt at a time, until you’re ready to tell it what to do. In essence, you’re training it before getting it to perform a task.

Giving it feedback

AI generators don’t actually know what a ‘good’ answer is; they’re just predicting the most relevant text word-by-word based on your prompt — check out the AI literacy page to learn more about how they do this.

If you don’t get the response you want, you can either rewrite your prompt or give it feedback. This technique is best when you can’t, or are not sure how best to, edit your prompt.

Give it feedback by telling it:

what it got right
what it got wrong
what you want instead

Remember to be SIC (specific, informative and clear).

In the following example, I asked the AI for ideas for building classroom culture in the first week of school with a new class.

Then, I gave it feedback telling it what I liked and what I didn't like, and why. I also gave it clear and specific instructions, reiterating key points I wanted to ensure it followed. The results second time around were much closer to what I was looking for.

Turn up the temperature

Ever feel like AI tools give predictable or unoriginal answers? This is because it generates its answers using a prediction based on its training data.

You can get it to give you more original and random answers by turning up the heat!

Text generators use temperature sampling, which is a scale of 0 (most precise) to 1 (most random). The closer the temperature is to 1, the less typical the answer will be. This is useful in two scenarios:

Prioritising factual accuracy — a low temperature is needed, e.g., 0.1.
Searching for original ideas — a high temperature is needed, e.g., 0.9.

The AI tool will automatically adjust the temperature based on your prompt, e.g., if you ask it to be creative. But you can specify the temperature by adding the following text to your prompt:

‘Use a temperature of __ to generate your answer.’

The example below follows on from the classroom culture ideas in the previous section. You can adjust the temperature, so if 0.9 is a bit too random, ask it to use a slightly lower temperature (e.g. 0.7).

Meta Prompts

Leverage the AI’s intelligence by asking it to come up with the best information or approaches and then feeding these back into the AI. This is particularly useful when you’re not an expert on the topic, want a different perspective or are looking for more insightful answers.

Tell it to play the role of an expert (or ask it which expert(s) are best suited) and give it some context, and then ask for approaches, frameworks or theories relating to your project. Choose the best and apply them to the task. You may wish to check it has a nuanced understanding of the approach, framework, etc., before moving on with the project.

This works particularly well for:

generating ideas
problem-solving
considering different viewpoints
subject matter expertise
evaluating your work

Check out the Meta Prompt blog post for more use cases and an example.

One step at a time

AI tools usually work best when long or complex tasks are broken down into steps.

Going step-by-step has a number of benefits:

Prompting is easier as you don’t need to write as many long prompts.
You can use insights gained from each step to inform the next one.
It’s easier to assess the AI’s reasoning to check its accuracy.

This approach may take longer, though, so probably isn’t needed for short or simple tasks. Also, AI generators have a limited memory (token limit), so they may start to ‘forget’ the start of very long chats.

How to do it:

Explain its role, the task and the context.
Give it the instructions for the step.
Go deeper, do further analysis and discuss until you’re ready to move on.
Repeat points 1 to 3 as necessary for each step.
Refer back to any great insights from previous steps and reiterate key information.

For complex tasks or long step-by-step instructions, try the phrases ‘let’s think step-by-step’ or ‘take a deep breath and work on this problem step-by-step’ right before your instructions. These phrases have been found to support an LLM's reasoning (Yang, et al., 2023).

Try the phrases ‘let’s think step-by-step’ or ‘take a deep breath and work on this problem step-by-step’ in your prompts for more complex tasks or instructions.

AI image generators

What to include in your prompts

Like with text generators, prompts still need to be specific, informative and clear. However, the way you prompt image generators, like Midjourney, is a little different.

Image generator prompts are usually short. Though, like speaking with a real artist, it helps to give specific and clear information about what you want the AI to create. Your prompt should include some or all of the following.

Setting or background

Specify the backdrop and surroundings, e.g., a serene beach or a bustling cityscape.

The subject and main elements

Define the critical components that are the focal point, e.g., a person, animal or an object, and describe them: what they’re doing, what they look like, etc.

Style / aesthetic

Mention the specific art style or technique, e.g., photo, retro or futuristic.

Framing and lighting

State how you want your subject to appear in the frame, e.g., close-up, portrait or low-angle. You can also specify the lighting, e.g., cinematic, studio or back lighting.

Here are some examples:

Prompt: 'office worker deep in thought, at desk with laptop, bustling office, photo, realistic, portrait'

Prompt: 'School kid, fighting giant textbook, comic style, wide shot'

Prompt: 'amused sloth in party hat and shades, close up, cartoon, disco ball lighting'

Tips for using AI image generators

AI image generators differ in quality and are advancing all the time. But, there are some general tips and techniques that can help you get better results.

Use descriptive language

Connect with your inner English teacher. Use specific synonyms that most accurately describe your image. Words like ‘happy’ or ‘sad’ are quite vague, so it’s better to use words like ‘amused’ or ‘frustrated’, whatever best describes what you’re looking for.

Sad school student sitting at desk, working, low angle, backlit

Frustrated school student sitting at desk, working, low angle, backlit

Edit and regenerate

Just like with AI text generators, try again if you don’t get the best result. You can either regenerate or tweak your prompt. Some image generators include editing features, such as inpainting, remixing or zoom out.

Use AI to help you write prompts

AI text generators can now read images. So, you can ask the AI to describe an image or style that’s similar to what you’re looking for, using this information to improve your prompt — avoid plagiarising, and cite other’s work when necessary.

I asked Bing Chat to describe an image similar to one I wanted to create — an adult reading to a child in Ancient Greece.

This information helped me write the prompt: 'old man reading to boy, contemplative, Ancient Greece, neoclassical', which created the image below.

Be multimodal

Some AI tools, like ChatGPT Plus, can generate text and images together. So with one prompt, you could generate a learning aid with images or a learning resource prototype. It still helps to include specific image details in your promtps.

Want to learn more about prompting?

Visit my blog for more prompting tips.

For interesting analysis and research summaries related to prompting, follow Ethan Mollick on LinkedIn, Twitter or his Substack: One Useful Thing.

To deep dive into prompt engineering techniques, check out Dr Lance Eliot on Forbes.

Although written for an old model of DALL.E, the DALL.E 2 prompt book by dall.ery gall.ery has loads of useful tips and examples to help you write image prompts.