Prompt engineering - getting the most out of your AI interactions

‘Prompting' = telling an AI what you want it to do for you. And while 'prompt engineering' may sound scary, it is just a collection of techniques to get the most out of your conversations with AI.

in this article...

  • Types of prompt
    • Conversation prompts 
    • System prompts
  • AI prompting tips and best practices
    • ✅ DO – Best practices
    • ❌ DON’Ts – Common pitfalls
    • 🎯 Structured Prompting (GCSE Method)
    • 🔄 Iterative refinement
    • 🌟 Advanced prompting tips for system prompts
  • Other prompt guides and resources
    • Prompt examples
  • Cautions and troubleshooting

Types of prompt

There are two broad categories of prompt.  

Conversation prompts 

Conversation prompts are the ones you use when chatting with an AI assistant - asking it questions or giving it instructions - e.g. ‘summarise this text for me’.

  • Each conversation you have is unique. 
  • In Theta Assist a conversation is called a thread.

This is where you enter your conversation prompt in Theta Assist:

System prompts

System prompts (sometimes also called project or developer prompts) are the prompts that tell an assistant what job it has, or how to behave across all the conversations it has. System prompts mean you don’t have restate your expectations every time you start a conversation - and can be a very powerful feature.

Simple system prompts

A system prompt can be as simple as telling the assistant to 'use a professional tone of voice and New Zealand spelling'.  Or setting the role perspective you want e.g. “You are an experienced science communicator, simplifying complex topics clearly.” 

Advanced system prompts

  • System prompts can be more advanced, and include multiple steps or instructions, output format and more.
  • For example - your prompt could say the assistant should 'act like an interviewer and ask the user a specific set of questions to gather required information, as follows...'.

System prompts in Theta Assist

In Theta Assist, you can edit the system prompt (also called 'instructions') when you create or edit an assistant.

 

AI prompting tips and best practices

✅ DO – Best practices

  • Be clear & specific – precisely state your objective.
  • Use examples – add examples directly in your prompts for clearer guidance.
  • Iterate & improve – continuously refine your prompts based on results.
  • Assign roles or personas – guide the model’s response style clearly.

e.g. “As a financial advisor, recommend conservative investment options.”

  • Use polite, easy-to-understand language – ensure your instructions are easily interpreted

❌ DON’Ts – Common pitfalls

  • Don’t overload information – avoid prompts that contain information that isn’t relevant to your task
  • Don’t use ambiguous negatives – phrase instructions positively and explicitly.

e.g. Replace “Don’t include irrelevant data” with “Include only essential data.”

  • Don’t expect complex tasks do be done all at once – split complex requests into simpler, smaller steps.
  • Don’t ignore limitations – set realistic expectations based on the chosen model’s capabilities.

🎯 Structured Prompting (GCSE Method)

Microsoft has created a prompt pattern which can be a useful way to structure your prompts:

  • Goal: Clearly define the task.
  • Context: Provide essential background information.
  • Source: Include any required references or materials.
  • Expectations: State your desired format, style, and tone.

Simple example:

“Write a 500-word renewable energy article (Goal) for beginners in environmental topics (Context), using 2023 International Energy Agency data (Source), with an informative yet engaging tone (Expectations).”

The same principles and pattern can be applied to more complex prompts.

Note:  It is perfectly fine to provide part of the prompt in one message, get the response from AI and then ask for refinements – like a conversation.

🔄 Iterative refinement

Refine prompts gradually via assessment and feedback:

  1. Start clearly and simply.
  2. Evaluate outcomes and clarify.
  3. Iterate with targeted follow-ups.

🌟 Advanced prompting tips for system prompts

  • Structured instructions: Use lists or bullet points.
  • Detailed context (when necessary): Provide enough detail so the model understands clearly.
  • Chain-of-thought: Request explanations step-by-step for complex tasks.

Adopting these practices will enhance your interaction, efficiency, and outcomes with large language models.

 

Other prompt guides and resources

  • https://platform.openai.com/docs/guides/prompt-engineering#six-strategies-for-getting-better-results
  • Beginners Prompting Guide
  • Prompt Engineering Guide - Google
  • Prompting Guide

Prompt examples

  • https://docs.anthropic.com/en/prompt-library/library

Cautions and troubleshooting

  • AI responses can be wrong - you should be comfortable asking AI to double check their thinking and or ask for evidence
  • You should always check critical information before sharing or using for key decisions.
  • AI models know a lot about the world, but they don’t (by default) have real time access to new things like today’s date or current affairs.
  • Try to understand the capabilities of the AI model you are using. Different models will respond differently. Some are better suited to particular tasks than others.
    • Factors:  Creativity, Context length
  • Even with the same prompts and the same model you will get different responses in every conversation. This is the nature of generative AI – an element of randomness is designed into each response.