Season 5: Innovating for impact

Make AI work for you, Kate Shepherd
Part 4 of 4

Crafting powerful prompts

Whatever you’re using AI for, it’s good practice to record the prompts you use, and the results given, to build a knowledge base or playbook.

By doing this, it will allow you to reuse the prompts, which saves time, and share them with others, which can increase consistency. For example, if you create a prompt that responds to customer emails in a certain tone of voice, it would be a good approach for everyone in your team to use the same prompts – helping you make the most of this innovative technology and saving time so you can focus on other innovations, like product development.

Get started by creating a shared document, spreadsheet or PowerPoint deck for you and your team, where each page contains the following information:

  • What the prompt does
  • What systems it needs
  • What data should be provided
  • The prompt
  • An example output
  • Any guidance as to when/where it should be used, also importantly when it should not be used
  • A rating out of 5 stars (i.e. how good do people think it is. They can provide improvement ideas too)

Building a better prompt for ChatGPT (or other Large Language Models – LLMs)

Artificial intelligence tools are sometimes concerned about how you write the prompt, its structure and syntax. However, there are more important reasons to have a better prompt syntax:

  • Easier for you to remember what it does, which is key if you want to re-use the prompt.
  • Easier structure for testing and extending its functionality.
  • Easier to transition from a conversational prompt to a GPT Bot instruction.

Prompt structure

Here are some guidelines to improve your prompt writing. Start by separating the sections of the prompt into 4 distinct parts:

  1. Pre-positioning: define how you are expecting the prompt to perform and detail any “persona” you wish it to follow
  2. Actions: the activity that the prompt is to follow
  3. Format: any format information or response guidance for how the output should appear
  4. Data: any data or information specifically for the prompt to access or act upon (or uploaded data files)

Within that, when we construct a prompt we use different brackets [ ] to suggest areas where the prompt could be varied, by enclosing the example within them. This makes the prompt easier to understand at a glance, especially if you were weren’t the person creating the prompt.

Here is an example for generating a to-do list:

Tip: to get to a new line in ChatGPT, without sending the command, press SHIFT Enter or SHIFT Return.

As an experienced executive assistant, you are going to organise my to-do list using the Eisenhower matrix method. [Urgent tasks have deadlines or target dates, Important includes working with other people, producing product or sales activity].

  1. Review the to-do list delimited by — to-do list —
  2. Organise the to-do list in accordance with the Eisenhower Matrix Method
  3. If it is possible for ChatGPT to help with any to-do list item suggest the appropriate ChatGPT prompt the user could use in a subsequent ChatGPT conversation

# Use the following format

The to-do item, follow this with a tick box

Priority: The Priority, Urgency: The Urgency
Estimate: The estimated time in minutes to do the task
Starter Prompt: an example prompt that ChatGPT could use to help with this task, denote user input required by {} include example within the {}

— to-do list —
[ paste your to-do list here ]

There are a few new techniques and tips introduced here.

  1. The use of “Steps” – we are using this to help to be clear to both ourselves and the AI that there are 3 specific actions we wish to have performed
  2. “Delimited by — to-do list —” – we are telling the AI that after this point in the prompt it is the data to be processed, it is not instructions. It also tells us, as users, where we should replace by pasting in our to-do list.
  3. We are asking ChatGPT to write future prompts which would be a starting point for completing those to-do items. These are often a good place to start, but probably as you become more skilful, you will instinctively improve them.
  4. “# Use the following format” and then specify the format. We are providing an example of how we would like the response to be formatted, a template of the response. The “#” symbol tells the AI that this is a new section of information, although still in the actions part of the prompt.

TRY THIS:
Here are two prompts, which you could take and improve as an exercise using the techniques above to improve structure, layout and re-use. For reference “bowtie risk management” is a system to assess risks and consequences.

Prompt 1
Using bowtie risk management, consider the top 3 risks that might occur in a warehouse that manually handles products that can weigh up to 15 kg each, there are also forklift trucks moving palleted goods.

Prompt 2
Do a web search on GDPR, what are the key points for UK businesses.

A final closing tip: if you install the app version of your chatbot (either Google, ChatGPT or Microsoft Copilot) it can also process images or photographs, so try taking a photograph of your to-do list (from your notebook) and ask for it to be transcribed, or even translated into another language.

Today is just the beginning of how AI can, and will be, used in modern businesses and everyday life. There is nothing to be afraid of- and the best time to start is now.

Thank you so much for joining this masterclass on making AI work for you. Best of luck on your journey integrating AI into your business.