How to write a good LLM prompt

A good prompt makes all the difference when interacting with an LLM. In today’s blog, we’ve broken down how prompts work, what ‘good’ really looks like and some prompt templates to get you started.

–

How do LLMs and prompts work?

To understand how to get the best output, it’s useful to know how LLMs work. An LLM (Large Language Model) is a type of AI trained on massive amounts of text to understand and generate human-like language.

It breaks text into tokens (small chunks like words or parts of words) and learns patterns between them from billions of examples. 

When you give an LLM a prompt, it predicts the next token again and again until a full answer is formed. So, the more data and parameters it has, the more nuanced, fluent, and context-aware its responses can be.

In short, AI is not a search engine; it’s a pattern matcher trained on huge data. As Arbor’s AI Engineer describes it: “AI is like a super-powerful autocomplete.” It generates likely responses based on how your prompt is framed, based on patterns, not facts or opinions. That means if your prompt is poor, the LLM’s response will also be poor. 

This also explains why hallucinations happen, as AI wants to answer your questions, so it will find the closest match to what it is you’re asking, even if that isn’t true. 

–

Three key points to remember

• Clearer inputs = better outputs
• The more specific you can be, the better – the more ways your prompt can be interpreted, the less likely you will get the answer that you have in your head
• Treat prompts like giving a task to an intern. The more context and clarity you give, the better the result

–

Anatomy of a good prompt checklist

We’ve put together a checklist of things that your prompt should include, or things you should consider when having a back and forth with an LLM.

• Goal: What do you want? (Summarise, write, explain, compare?)
• Context: What background is relevant?
• Format: List, paragraph, table, steps?
• Constraints: audience, style
• Timeframe: data timeframe
• Tone (optional): Professional, friendly, concise etc,
• Objectivity: Avoid adjectives like “wins/losses, best/worst, effective/ineffective, large/small, high/low, broad/narrow, significant/insignificant, Improved/deteriorated, reliable, simple, intuitive, efficient, robust, etc. The AI does not have context into your baseline. 
• Avoid leading questions: “Isn’t it better to…” 
• Always be specific 
• Speak in whole sentences and explain what you want
• Give the instruction up front, then repeat it at the end of the prompt 
• Be firm. You don’t need to be polite!

–

What does a good LLM prompt look like in practice?

So you’ve got your building blocks, here’s some ‘before’ and ‘afters’ to help you gauge what a good prompt looks like in reality. 

Original Prompt	Improved Prompt
“What’s our AI policy?”	“Summarise our AI policy in plain language for a new hire who just joined.”
“Help with interventions”	“Explain how to create and carry out an interventions strategy for a student who is persistently absent.”
“How do I do a seating plan?”	“Outline the key steps in creating a seating plan for a class of 30 children.”

–

LLM prompt templates

Here are some templates to get you started:

1. General information retrieval

Examples:

• “Summarise the key points of the Ofsted inspection framework for secondary schools.”
• “Explain the responsibilities of a school governor in simple terms for new board members.”
• “What are the main safeguarding obligations for a Designated Safeguarding Lead in the UK?”

Template: “Explain [policy/procedure] in simple terms for [role or audience], based on [official framework or guidance].”

–

2. Content creation

Examples:

• “Write a parent newsletter summarising upcoming school events for the autumn term.”
• “Draft a welcome letter for new Year 7 students and their families.”
• “Create a sample lesson plan on climate change for KS3 geography following the national curriculum.”

Template: “Write a [format] about [topic], using a [tone], for [audience], based on [curriculum/framework].”

–

3. Summarising & finding trends in data

Examples:

• “Summarise trends from the Year 10 attendance records over the last 3 terms.”
•  “Identify concerning behavioural patterns based on incident logs from this term.”
• “Summarise the top concerns raised in the parent feedback survey from summer term.”

Template: “Summarise trends in [data/report] from [timeframe], highlighting [insights or anomalies].”

–

4. Troubleshooting / guidance

Examples:

• “How do I report a safeguarding concern using CPOMS?”
• “What should I do if a parent refuses to pick up their child who is unwell?”
• “How do I escalate a persistent lateness issue under the school’s attendance policy?”

Template: “What steps should a [role] take to resolve [problem], following [policy/system/tool]?”

–

5. Curriculum & Lesson Support

Examples:

• “Suggest differentiated reading activities for a mixed-ability Year 8 English class.”
•  “How can I adapt a science lesson on ecosystems for a student with ASD?”
•  “Provide three starter activities for a Year 5 maths lesson on fractions.”

Template: “Suggest [number] [activity type] for [subject/topic], tailored for [year group or need].”

–

6. Communication and behaviour management

Examples:

• “Write an email to parents about a student’s ongoing behavioural concerns in a respectful tone.”
• “Draft a behaviour support plan for a pupil at risk of exclusion.”
• “Provide strategies for de-escalating classroom conflict between two pupils.”

Template: “Draft a [document type] addressing [issue] with a [tone/approach] for [audience].”

–

7. Events and logistics

Examples:

• “Create a checklist for planning a Year 6 residential trip to Wales.”
• “Outline the key roles and timeline for organising a school open evening.”
• “Draft a risk assessment for a school trip to the Science Museum.”

Template: “Outline a [checklist/plan/timeline] for [event/task], including key responsibilities and deadlines.”

–

More resources on AI best practice

Hear from experts in the Arbor community and beyond in our on-demand webinars, reports and on Arbor HQ