Open Immersive Reader

When you first open up Copilot for example, you are faced with an empty message box. This can actually be quite daunting at first – you obviously have an objective in mind, something that you want to achieve – but how do you ensure that you successfully interact with this piece of technology so that you get the desired outcome?

The key is with the prompt.

Imagine that the interface is simply a chat with a colleague, perhaps imagine that it is a conversation between you and the newest member on your team. How would you interact with them? Would you simply say “generate me a report on last months performance against KPI”, or would you be polite, and provide context in terms of where to find the report files, and what to do with them?

In general, you get out what you put in. If you give it an ambiguous prompt that is lacking context, you will likely get incorrect or insufficient information back. In this article we break down the key components of a good prompt, to hopefully inspire you to get creative and operate LLMs more efficiently.

Let’s use Microsoft 365 Copilot as an example here.

1. Start with the objective in mind

Tell it what you want to achieve:

2. Set the scene

This is where you give Context to the LLM to help it hone in on exactly what you need.

3. Define the scope

Now, you tell it how far to go, what files or areas to look in, and give it a timeframe so it can apply the context that you gave it in step 2.

4. Adjust the end result

Here is where you can tell it the intended audience, so it can tailor the response accordingly. This could be to make it short or verbose, or use detailed language for a technical audience.

5. Manners, Curiosity, and finally – Persistence

Always use basic polite language and etiquette when interacting with LLMs. Generative AI will learn from you, and mirror your intent, so to avoid any issues it is always best to be as professional and clear as possible.

If your prompts don’t seem to be working, you need to persist. Keep asking it to adjust things, and iterate on its responses. Keep the conversation going, and you should eventually end up meeting your initial objectives.

6. Your own personal Prompt Engineer?

If all else fails, and after the umteenth attempt you are still not getting anywhere, one final thing to try is to use the LLM to craft the prompt to achieve your end goal for you. Yes, seriously!

You can use the same format as above, but preface it with it’s own specific scenario, telling it how to act and what to do. Here’s an example that I used to help me write another blog post:

My Prompt

The Response

In Summary

Prompting is an art, and something which you will become more effective at the more you use it. The trick is to not get discouraged, and to find other ways of achieving what you need it to. Iteration is a virtue.

Written by Louis Arthur-Brown, Head of Cloud and Managed Services, Codestone

Contact [email protected] to talk to our experts.