Sometimes, you will need a quick and easy way to check for your pupil's understanding after you set learning homework. Even though multiple-choice quizzes have their limitations, they can provide you with a quick icebreaker to set the mood of the lesson. This type of activity may require a bit more iteration than those presented in Level 1, but once you've got a formula that works for you, you will be able to use it again and again to create other quizzes.

At the moment of writing this book, most large language models struggle to create long quizzes of more than 20 questions, and they get to a point where they start either repeating themselves or making up things (please see the chapter discussing hallucinations). Try to be as specific as you can in your prompt to get the best results. Consider: How many questions do you want? How many options per question? How many correct answers? What is the type of options and the language to be used? Also, which elements need to be present both in questions and in the options?

Like in the previous ideas or prompts, you can specify the level of your class, the subject, the topic, and the specific elements you want to be tested. You can also state the purpose of the quiz and the learning objectives.

Fill-in-the-gaps activities are excellent to assess your pupils' understanding of key concepts or vocabulary and, crucially in foreign language learning, to assess their grammatical knowledge. With AI, this type of activities can be created in just a few seconds if we know how to ask effectively. I've included this idea in Level 2 because you most certainly won't get the result you want the first time around if your prompt is not specific enough. If you simply ask the large language model to create a filling-the-gaps activity on topic X, Y, or Z, you may be disappointed. Instead, provide the chatbot with a comprehensive list of instructions stating exactly what you want your filling-the-gaps activity to be like.

The first thing you need to do is to determine whether you will want the model to generate a text or if you will provide one of your own. If you need it to create a text, you will need to be as specific as possible about the nature of the text (you can refer to the chapter Level 1.3. Text generation for this). For the fill-in-the-gaps part, you will need to give instructions as to: how many gaps would you like the text to have; whether you want them to be numbered; whether you want a space to be included for your pupils to provide an answer; the words or type of words that you would like to be excluded, etc. And as ever, you can refine even further by stating the purpose of the activity or what the learning objectives are.

Similarly to the previous example, match-up activities require some iteration due to most large language models with today's technology sometimes struggling to randomise, sort, or organise columns with elements that match another column without repeating or missing any elements. However, if done successfully, a match-up activity is an excellent icebreaker or scaffolding task to work on literacy by practising vocabulary and definitions, or any other uses you may think of.

The simplest match-up activity would be words and definitions or, in the case of modern languages, words and translations. As ever, you can either provide the large language model with your own list, or you can ask it to create its own list of words based on the theme and topic of your choice.

You will have noticed that sometimes the objectives presented in syllabus or course books are not specific enough, or really measurable, or they lack any other of the qualities specified in the acronym SMART: Specific, Measurable, Achievable, Relevant, and Time-bound. You can use a large language model to transform your objectives so they meet all the criteria in the SMART acronym.

However, this is not as easy as to ask the chatbot to turn or to make your objectives SMART, and on a few occasions, I've had a large language model listing all the letters of the acronym and giving separate objectives for each of the letters. So, what type of magic prompt do you have to use to make your objectives smarter, you may be wondering? I would say that the first rule would be to not assume any previous knowledge in the chatbot and to not take anything for granted because even the more advanced reasoning models can fall into the trap of listing the words in the acronym and providing a separate objective for each of them.

So, what I will do here is suggest a prompt to which you will have to add the objectives you would like to modify, and with the output you get from the large language model, you will need to do some editing or follow-up requests.

There have been many instances in my teaching career where my adaptive teaching has suffered due to time constraints, and oftentimes, it's virtually impossible to cater for all types of learning and all types of students for every single one of your lessons. With large language models, we have now the possibility to do this seamlessly by using the right prompts; and I'm not talking only about texts here, but all types of learning materials, from more simple activities to more complex worksheets, rubrics, or even past exam papers.

Everything can be adapted to your students' specific needs or learning styles. There are some caveats we need to consider and strategies to use in your prompts to achieve this. If you have been reading this book in a linear way, you are probably thinking already about all the specificities you need to include in your prompt for it to be effective; but crucially, defining the audience and the purpose is here more important than ever, no matter if we are talking about a specific activity, a text, a worksheet, a past paper, or any of the other elements discussed earlier.

LLM’s response:

Got it — you want the same storyline (the jewel thief) but adapted to Year 9 so that: Easy = all can complete confidently, Medium = most can complete, with some challenge, Hard = stretching, with more steps or trickier contexts so only about half get full marks. Here’s the adapted activity with three tiers. I’ve kept the Pythagoras theme but scaled the numbers, diagrams, and problem complexity.

Easy Level

(Confidence builders – straightforward right triangles, whole numbers where possible). For each question, round to 1 decimal place if necessary.

1. The thief runs 3 m north, then 4 m east. How far is the straight-line distance from the start to the finish?
2. A right-angled triangle has one side of 5 m and another of 12 m. Find the hypotenuse.
3. A ramp is 10 m long and reaches a height of 6 m. How far is it from the bottom of the ramp to the wall?
4. The thief climbs a ladder 5 m long, placed so that its base is 3 m away from the wall. How high up the wall does it reach?
5. A flagpole is supported by a rope from the top to the ground 5 m away from its base. If the rope is 13 m long, how tall is the flagpole?

Medium Level

(Less neat numbers, mixed finding hypotenuse and shorter side, adding context steps)

1. The thief runs 15 m across a park, turns at a right angle and runs 8 m. What is the straight-line distance from start to finish?
2. A right-angled triangle has a hypotenuse of 10.2 m and one shorter side of 4.5 m. Find the length of the other side.
3. A 4 m ladder leans against a wall with its base 1.5 m from the wall. How high up the wall does the ladder reach?
4. The thief hides in a building with a square floor plan, 12 m on each side. He sneaks from one corner directly to the opposite corner. How far does he travel?
5. An isosceles triangle has equal sides of 13 m and a base of 10 m. Find its height and then its perimeter.

Hard Level

(Messier decimals, multi-step problems, and real-life twists — requires careful organisation of working)

1. The thief rides a motorbike 22.4 m north, then 37.8 m east. Calculate the straight-line distance back to the start.
2. A right-angled triangle has a hypotenuse of 15.7 m and one shorter side of 9.8 m. Find the other side, to 1 decimal place.
3. A ladder 7.5 m long leans against a wall, with its base 2.2 m away from the wall. How high does it reach?
4. The thief escapes to a rooftop garden shaped like a rectangle, 18 m by 7 m. He runs from one corner to the opposite corner, but halfway across he drops a bag and must run directly to it. The bag is 12 m along the diagonal from the starting point. How far does he still have to run to reach the far corner?
5. An isosceles triangle has a base of 8.6 m and a perimeter of 26.4 m. Use Pythagoras to find its height.

If you like, I can add diagrams for each problem so the “clues” feel like an actual detective trail with visual prompts — that will make it much more engaging and accessible for Year 9.

I did ask for a diagram and this was the model’s response:

Here’s the first diagram for Easy Q1 — a simple 3–4 right triangle for pupils to find the hypotenuse. I can now produce similar labelled diagrams for all questions in easy, medium, and hard versions so the jewel thief’s “trail” is fully illustrated. Would you like me to make all 15 diagrams in one set for you to use?

In the beginners level, I have discussed how to use large language models to help you brainstorm ideas. In this section, I will expand on this idea in the context of project-based learning.

Project-based learning is a pedagogical approach where students are given a goal or final task that they need to execute in the form of a project, but to be successful, they need to develop their skills and knowledge in one or several areas in such a way that the relevance of the project lies within the process itself rather than the final product. Sometimes, project-based learning can take a multidisciplinary approach where pupils need to draw knowledge and skills from several subjects to be able to successfully complete their projects.

From a teacher's perspective, to design something like this is a huge task, and sometimes it even involves several teachers from different departments. Large language models can help you streamline this task by speeding up the brainstorming process and crafting all the necessary materials for your pupils to develop their relevant skills. It can also craft the SMART objectives of the project for you and even the rubric with the success criteria. (Please see the relevant sections in this book.)

The number of approaches and examples is too large to be encapsulated in a single prompt, but I will guide you through the steps I would take to craft a project-based learning experience for my pupils.

1. The first step is to define the purpose of the project. This will naturally be based on your syllabus or scheme of work and will be linked to the learning outcomes. Once you know this, you can use a large language model to define the SMART objectives, along the lines of: "You are an educational designer, an expert in the subject ‘X’. I need you to help me craft SMART objectives for my project-based learning experience based on the following outcomes." And then refer to my instructions on how to craft SMART objectives in the beginners section of this book.
2. Now that you've got the SMART objectives, use the same chat window to brainstorm project ideas. The large language model will remember the context of the conversation if you are using the same window, so you don't need to repeat yourself. Therefore, you can simply ask: "Please provide 10 ideas for a project-based learning experience based on the learning objectives and outcomes.”
3. Simply read through the ideas or examples in the large language model output, and if you don't like any of them, you can simply iterate or refine until you find the perfect idea for your class. To refine, don't simply ask the large language model to repeat the process of giving you 10 ideas again, but perhaps select one or two ideas that you liked and ask it to provide five more examples along those lines.
4. Once you have selected the idea for your project, you can ask the large language model to list all the scaffolding activities that your pupils will need to develop their skills and knowledge to successfully finish their projects. Once again, select the ones you like the most, ask the large language model to give you some more if necessary.
5. When you are happy with the number of tasks or activities for your scaffolding, you can tackle them one at a time and ask the chatbot (still using the same chat window) to craft the materials required for the project using the prompting guidance that you can find in this book.

I have omitted here the rest of my conversation with the model, but you can have a look at all the follow ups and responses in my video tutorial ‘2.6. Generate Project-Based Learning Experiences’, where I ask the AI to write me a teacher planner and a student’s booklet for the project.

Another way to leverage AI to help you develop your materials is to ask it to create a clear set of instructions and assessment criteria for any given task. Unlike some of the uses seen in Level 1, you will need to be quite specific to get results of any quality.

Step 1. Define the role, the goal, the audience, the subject, and the topic, along with what you would like the large language model to create for you. For example: "Act as a teacher of Spanish. I need you to create a rubric for an activity my Year 9 class will do on the topic daily routine with the objective of learning the reflexive verbs."

Step 2. Detail the description. If you already have the task for the rubric you want to create, simply feed it to the large language model after the previous prompt. If not, you will need to describe in detail the task that you have in mind, remembering that the more specific you are, the more accurate the large language model output will be. For example, describe a task like: "A 120-word writing on the topic daily routine with 4 bullet points that the students will need to cover in the form of an email. For example, they will need to use 3 time frames (present, past, and future), at least 5 reflexive verbs, connectives, adjectives, and verbs to express opinions."

Step 3. Iteration. Due to the complexity of the prompt and the task, there is a good chance that the output of the large language model will not be exactly what you are looking for. Similarly to other uses of large language models, you will need to follow up with refined requests to get the best possible rubric for your class and the task you want them to complete.