How to Write Prompts: an Easy Guide for Teachers

Here’s an easy way to assess the effectiveness of your prompt – the text that goes into the large language model (LLM): Ask yourself if your students will be able to successfully complete the task you’ve assigned to the language model.

A) If the answer is ‘Yes’, then you’re good to go – proceed with confidence.

B) If the answer is ‘No’, review your prompt and make adjustments.

If you’ve honed your skills in classroom management and lesson planning, especially for young learners and teenagers, and are able to elicit the desired behavior from your students, then you’ve already grasped the core principles of crafting an effective prompt for a large language model (LLM).

A few days ago, I attended Dr. Amin Davoodi’s presentation at the #ElliiCon2023 conference. The topic was ‘Practical Strategies for Empowering Language Learning with Humanistic Approaches to Technology and AI.’ I must say, the presentation was exceptional. Unlike the typical ‘Look at what it can do’ or ‘Look at what I can do with it’ approach, the presenter began with a solid foundation – why we need AI, supported by thorough research. Throughout the presentation, he shared numerous examples and practical strategies for effectively integrating AI into the classroom.

However, what truly struck a chord with me was the key message: ‘AI only magnifies who you are as a teacher.’ 

This can’t be truer than true.

The good and responsible use of AI in education is rooted in our teaching approach and expertise. AI, at times, can seem almost magical, leading people to believe that the tool itself is the key. This perception has opened the door for countless entrepreneurs who think they can create educational technology simply because they have a basic familiarity with the field, often driven by the notion, ‘I’ve been a student, so I know how it works.’ This explains why we see an abundance of platforms emerging, all claiming to ‘save you time’ by mimicking what we’ve been doing in education for years. However, as the saying goes, ‘garbage in, garbage out.’ The true quality and impact of the educational content depend not merely on the existence of technology but on the expertise and intention behind the technology.

It’s often said that magic is a carefully planned moment. To achieve quality results, two essential components are required: a) a clear understanding of what you aim to achieve, and b) the ability to evaluate whether you’ve achieved it. If a teacher lacks experience in materials writing, enrolling in a course on using AI to create materials won’t miraculously transform them into a skilled materials writer. Similarly, if a teacher hasn’t previously developed their own courses or curricula, they won’t suddenly produce a top-notch course or curriculum, whether AI is involved or not. 

In essence, your expertise is the foundation that will be put into a prompt that guides the model to get the result you want.

The good news is, if you’re a teacher, especially one who’s taught teenagers and has a background in linguistics, you likely already know how to create a good prompt.

Now, let’s go over the key points:

Setting context

Getting to know teenagers is a key part of teaching them. We should understand their personalities, life situations, interests, and what drives their actions – this is what we mean by ‘context’. It helps us connect with them – build rapport – and make learning happen.

Likewise, when working with a language model, we need to set context by providing it with the relevant information. This background info helps the model give accurate responses. Without it, the model will just spit out Barnum statements or completely irrelevant responses.

Unlike teenagers, who often respond with a vague ‘I don’t know’ for various reasons, sometimes just to avoid answering, the model never uses this response. It’s designed to answer questions around the clock. The model’s behavior is similar to when you ask your teen students to do their homework, and they complete it just to check it off the list, without much (well, any) thought or effort.

So, when your prompt lacks clarity, context, or reference information, the model adopts the ‘human nature in its honest and naked form’ and starts producing ‘satisfactory bullshit’.

Give clear instructions

Providing clear instructions is vital to ensure that students fully understand what is expected of them. Consider this scenario: a teacher assigns a 250-word essay on the topic ‘Environment’ without offering any further guidance. Now, what are the odds that students will produce a high-quality essay? Well, ‘dum spiro spero’ aka hope is a powerful motivator, but realistically, the chances are quite low, very low.

When formulating tasks for students, it’s crucial to furnish them with comprehensive instructions that outline precisely what you expect from them, including what sources to consult, how to structure their essay, which format to use, and so on. Often, it’s beneficial to provide examples that they can follow as well. 

Similarly, it’s essential to provide VERY clear instructions to the model, specifying precisely what you want it to do and outlining the step-by-step process it should follow. Improve clarity by using bullet points to list your instructions. In some cases, repeat the key word twice.  

Your instructions should also consider the model’s capabilities. For instance, asking the model to generate a 100-word text indicates you want a relatively short paragraph. However, expecting the model to produce exactly 100 words can be unrealistic since large language models (LLMs) don’t perceive words; they operate in tokens.

So, if, for example, you want a sentence composed solely of words beginning with ‘th’ (like in the TH-Sound Board), you’ll need to work out how to convey this specific requirement to the model effectively.

Instead of telling a young learner or teenager what they shouldn’t do, like ‘stop talking’ or ‘don’t check the answers early,’ that often (always?) encourages them to continue their behavior, it’s better to tell them what they should do. It’s the same with the model – give it clear, positive instructions for better results.


Just like with teenagers, it’s best to keep things simple and clear for the model. Avoid assuming it ‘knows’ a lot about any specific topic. Simple language and examples work best to get the results you want, just as you’d simplify instructions for your students to get better responses.

Provide reference information

If you want to get reliable outcomes, it is important to provide some ‘source-spiration’ to the model. This helps prevent bias and ensures that the outcomes align with our expectations. Just like you would guide teenagers with reliable sources, providing reference information to the model improves the quality of its responses.

Set tone

Teenagers often struggle with their emotions, and research says they find it hard to tell different feelings apart. For instance, in a test, grown-ups could look at a face and correctly say if the person looked scared, shocked, or angry. But only about half of teenagers got it right.

When you talk to a big language model, just be clear about the kind of answer you want. Say if you want it to sound nice, informative, or a little funny. This way, you’ll get the response you’re looking for. If the answer seems dull, it’s probably because you didn’t say what tone you wanted.

Just like how teenagers adjust their behavior in different situations, the model can change its writing style to match different needs. Teens act differently with teachers, friends, and in other situations. Similarly, the model can mimic the tone of a formal expert, a friendly conversationalist, or even a humorous character to fit different writing tasks.

A final thought

Teenagers can change their minds, and so can the model. You can’t always have complete control over the outcome in either case. AI might be unpredictable and not always give you the results you expect, and your strategies might not work perfectly the first time. This means that each time you use a prompt, you could get a different outcome. It may take some time and iterations to get the desired output.

However, just as with teens, don’t be discouraged by initial hiccups. Much like how constructive feedback helps teenagers learn, it’s essential to give feedback to the model to refine its responses. It’s a bit like life – things can vary, and you have to be flexible. Experiment with different prompts and approaches to find what works best.

Happy teaching!


Leave a Reply