Time-saving conversations with LLMs

Learn the prompting skills you need to reduce workload

Alan Harrison

Practical strategies for using free AI chatbots to reduce teacher workload through effective prompting frameworks.

31% of UK teachers have not used generative AI (GAI) for school work. Meanwhile, 38% of UK schools did nothing in the last year to reduce workload (Teacher Tapp, August 2024). Why not kill two birds with one stone: tackle workload by harnessing the power of AI? While there are plenty of edu-focused tools out there that provide a friendly user interface for specific tasks – think Teachmate, Curipod, MagicSchool, Brisk and many more – most come at a financial or opportunity cost, or both. But you can do a great deal with trusty, free chatbots such as ChatGPT, Copilot, Claude and Gemini, simply by chatting with them: no great learning curve or investment in software needed. Maybe you tried the bots before and were left unimpressed. But we know much more about prompting a large language model (LLM) such as ChatGPT than we did during the LLM explosion of 2023. Let’s see how to up our prompting game and get better results from free chatbots.

Prompting with purpose

Dan Fitzpatrick, known as “The AI Educator “, created the PREP framework to help teachers get better results from LLMs by structuring their prompts. Let’s explore PREP now:

• Prompt – this is what you want the LLM to do, for example, “create a plan for a lesson on prokaryotic and eukaryotic cells”, or “write a fictional dialogue between Macbeth and Duncan where Duncan asks Macbeth why he killed him.”
• Role – define the character you want the LLM to adopt. For example, “you are an experienced teacher of secondary science who is passionate about biology”, or “you are alternately King Duncan and Macbeth in Shakespeare’s Macbeth!”
• Explicit instructions – give guiding context here. LLMs have been trained on literally trillions of words: a prompt including context such as the specification, pedagogy and reference answers will trigger more helpful patterns. For example, “include retrieval practice questions, checks for understanding and a plenary quiz”, “follow the AQA Geography specification, assume this response _____ is a model answer”, or “write engagingly in the style of a dialogue to be used as a stimulus in an English lesson.”
• Parameters – here you can include layout, mood, language, reading age and much more: “for Year 7 mixed ability in England, single page, each lesson phase has heading and up to 100 words”, “questions under 50 words, mark scheme laid out similar to this example ____”, or “write for Year 7, max 15 words per sentence and max 250 altogether. Provide teacher notes explaining why the characters say what they do, with references to the original play.”

AI is a librarian not a teacher

Before we dive into more examples, it’s important to acknowledge that while AI can generate lesson plans, write exam questions, and provide discussion prompts, it doesn’t “understand” content the way a teacher does. AI works by predicting the most likely next word based on its training data—it doesn’t “know” Shakespeare, nor can it reason in the way a human can. Caused by these issues, some of the pitfalls are:

• misinformation & “hallucinations” – AI can generate convincing but incorrect content. Always double-check factual accuracy.
• bias in responses – AI reflects the data it was trained on, which can sometimes introduce unintended biases.
• over-reliance on AI – it’s tempting to let AI do all the heavy lifting, but it should enhance, not replace, professional judgment. As the teacher, you remain the expert in the process.
• student misuse – if AI can generate an essay plan, so can students! Teaching them how to use AI ethically is crucial. Have an AI usage policy and communicate it often.

All of this means that teachers must retain their autonomy over pedagogical decisions, in short: outsource your “doing” not your “thinking.”

More prompt ideas

We will have a look at the Macbeth prompt in action shortly, but here are some more “PREP” structured prompts for other subjects:

• GCSE Science: “You are an experienced science teacher. Generate five GCSE biology multiple-choice questions on osmosis, following the OCR specification. Include a mark scheme, explanations for correct answers and teacher notes on potential misconceptions.”
• Year 8 Maths: “Create three Pythagoras’ Theorem questions for Year 8: basic, intermediate, and challenging. You are an experienced secondary maths teacher who scaffolds learning effectively. Provide step-by-step solutions. Keep each question under 50 words. Align with UK Year 8 standards. Use correct notation and clear reasoning.”
• Year 3 History: “Create a simple, engaging timeline activity for Year 3 students on the Ancient Egyptians. You are a primary history subject lead who makes learning interactive and fun for 7-8-year-olds. Include 5-7 key moments (e.g., pyramids built, Tutankhamun’s reign). Use clear, child-friendly language and keep instructions simple (reading age 7-8). Provide a printable worksheet and timeline cards. The activity should take 20-30 minutes and work for mixed abilities.”

PREP in action

So, does it work? Here is our complete “PREP” structured prompt which we will put into the free version of ChatGPT. I am using the free version, no need for any subscription fees.

“Write a fictional dialogue between Macbeth and Duncan where Duncan asks Macbeth why he killed him. You are alternately King Duncan and Macbeth, Thane of Glamis in Shakespeare’s Macbeth! Write engagingly and clearly in the style of a dialogue script for two people, the result to be used as a stimulus in an English lesson. Write for Year 7 (age 11-12), maximum 15 words per sentence and a maximum of 250 words for the whole dialogue. Provide teacher notes beneath explaining why the characters say what they do, with references to the original play.”

This is the result, and for maximum effect, I have also illustrated it using ChatGPT (although GAI image generators are still very bad at fingers and text!)

Macbeth and Duncan: A Ghostly Confrontation

Scene: A dark, misty space. Duncan’s ghost appears before Macbeth.

DUNCAN: Macbeth! Why did you strike me down? I trusted you with my life!

MACBETH: My lord, ambition whispered in my ear. The crown was within reach, and I could not resist.

DUNCAN: Was my kindness not enough? I honoured you, made you Thane of Cawdor!

MACBETH: Aye, and with that title came hope. Hope that I might wear the crown myself.

DUNCAN: Hope built on murder? Did you not fear justice? Or heaven’s wrath?

MACBETH: I feared everything! But fear drives men to act. The witches spoke, and their words burned in me.

DUNCAN: Their words! Did you not have a choice? Could you not have let fate decide?

MACBETH: Fate? I thought I could shape it. I wanted to be king, not wait for chance.

DUNCAN: And now? Is your throne safe? Your hands clean?

MACBETH: Blood stains my soul, my sleep, my every thought. There is no peace for me now.

DUNCAN: Then was it worth it?

MACBETH: No. A crown of gold, but a heart of stone. I have lost myself.

Teacher Notes

This dialogue explores Macbeth’s motivations and guilt, referencing key themes from Shakespeare’s play.

• Duncan’s Trust: Duncan’s line “I trusted you with my life” recalls his misplaced faith in Macbeth (Act 1, Scene 4).
• Ambition’s Power: Macbeth justifies his actions by blaming ambition, reflecting Lady Macbeth’s manipulation and his own desires (Act 1, Scene 7).
• The Witches’ Influence: Macbeth refers to the witches’ prophecy, showing how it planted dangerous thoughts in his mind (Act 1, Scene 3).
• Fear as a Motivator: Macbeth claims fear drove him to act, linking to his paranoia and eventual downfall (Act 3, Scene 4).
• Fate vs. Free Will: Duncan questions whether Macbeth could have let fate take its course, echoing the play’s debate on destiny (Act 1, Scene 3).
• Macbeth’s Guilt: His final words highlight regret, mirroring his later soliloquies where he laments his crimes (Act 5, Scene 5).

PREP to PREPARED

While this is undoubtedly usable by an experienced teacher, we don’t have to stop there. ChatGPT is a chatbot, so let’s chat with it. Maybe you’re wondering how to use this in the classroom? Just ask “give me some lesson ideas for using this in the Year 7 English classroom.” We can also go back to Dan Fitzpatrick and try the “ARED” steps he added to the PREP framework to create PREPARED:

• Ask – Prompt the LLM to evaluate its response for accuracy and clarity. For example, “How could this be improved for better engagement or accuracy?” Evidence shows that asking LLMs to critically assess their output results in more refined, relevant, and coherent responses.
• Rate – Ask it to rate the quality of its response, for example, “How would you rate this response out of 10 where 10 is highly useful, pedagogically sound and very engaging, explain your rating and suggest how it can be improved.” This prompts an evidence-based self-assessment that improves accuracy.
• Emotions – appeal to the bot’s apparent “emotional” side. It might sound odd, but AI sometimes gives better responses when you add urgency or emphasis. This is because LLMs have been trained on human conversations, where people often emphasize importance before giving detailed answers. Try: “This is very important, I’m presenting to senior leadership tomorrow”, or “Take a deep breath and focus. Make sure this answer is clear and pedagogically sound.” This technique doesn’t actually trigger any emotional response, it simply activates patterns in the training data where human-like responses become more precise.
• Diversity – we talked earlier about bias, and due to popular demand, Dan Fitzpatrick recently added “D” for Diversity to his framework. We could ask questions like “are there any underrepresented perspectives or potential biases in this content, and how can we make it more representative of all viewpoints?” This may create resources that help students feel better represented and valued.

Backwards mapping

Another approach you can take with LLMs is known as “backwards mapping.” Take an exemplar response (or ask ChatGPT to create one for you) and then ask it to generate a lesson based on that response. This works great if you already have an example of “what a good one looks like” (WAGOLL) such as a long-answer question response to which you previously awarded full marks. Just ask the bot how you would go about teaching to achieve the same result in one lesson. For example: “This is a model answer to a question on the impact of the Norman conquest ______. Create a lesson plan to achieve this result in a Year 8 class.” Don’t forget to keep chatting, using PREPARE to get incrementally better results until you are happy!

Next steps

I hope this journey through effective prompting has been illuminating. So, where do you start? Try one of the prompts above and see what you get. Then, refine your results using PREPARED. Now share your best prompts with colleagues and build a school-wide AI resource bank. Want inspiration? Ask ChatGPT, “I teach [subject] to [year group]. What are some ways I can use AI to save time?” Just maybe, you’ll be pleasantly surprised.

References:

• PREPARE framework courtesy of Dan Fitzpatrick at io
• The entire Macbeth chat, created by following the PREPARE dialogue can be found in the saved chat dialogue here: ly/macbethduncan.