Write Better Prompts with a Prompt Card

Prompt Card

As AI continues to improve and become integrated into more of the IT tools we use in our daily work and private lives, the ability to write effective prompts becomes increasingly essential, allowing us to leverage these potent tools to our advantage.

Admittedly, it can sometimes be challenging to write clear, precise, and well-structured prompts. This is where the prompt card comes into play. A prompt card is a template that guides us through the process of writing structured and effective prompts based on best practices and an understanding of how language models work, ensuring we get the highest quality responses.

I have created this prompt card for my teaching on how to write professional prompts, and I also use it myself when prompting. Here you can see an example of how I have filled out the prompt card with the first version of a prompt to analyze the result of an A/B test:

Promptkort eksempel analyse af ABtest

How to Fill Out the Prompt Card

The idea is that you print it out and fill it in by hand, as this provides better overview and structure than the small input fields of AI models. It is often a good idea to fill it out together with your colleagues, as many tasks are interdisciplinary and require a comprehensive approach to the issue.

The prompt card consists of four fields, with four different instructions that should be filled out.

1. Tell Your AI Which Expert to Be

First, tell your AI what expert role you want it to assume to complete the task you set it. In the example, I instruct ChatGPT-4 to be an expert in conversion optimisation, that is, a person specialised in optimising online businesses such as webshops or apps.

Then define the competencies and knowledge it should have. In the example, I have summarised this in a list of concepts, tools, and analysis methods necessary to analyse an A/B test and participate in the subsequent optimisation process, including: A/B testing, web analytics, visitor statistics, conversions, conversion rate, e-commerce, web design, usability, and behavioural psychology.

Language models like ChatGPT, Gemini, and Copilot are, broadly speaking, super advanced autocomplete algorithms, like those you know from your phone or Word. They predict the next word based on the current text. By providing a detailed description of your AI’s role, skills, and knowledge, you help it predict and generate more precise and relevant responses.

2. Explain the Task to Your AI

Now it’s time to tell your AI what task it should perform for you and how it should do it. Think of this part of the prompt as a recipe the AI must follow, just as you would explain to another person how to bake bread, do the dishes, or use the remote control for your TV.

Break down the task into well-defined sub-tasks for your AI to solve one at a time. This makes it easier for your AI to follow instructions and produce the desired result.

In the example, ChatGPT’s task is to analyse an A/B test, write a short and easy-to-understand analysis that starts with the conclusion, focuses on whether the result is positive, negative, or neutral, and is reliable. This is a first draft of a prompt.

3. Provide Your AI with Necessary Background Information

Give your AI the necessary background information it needs to complete the task. This could include information about you, your company, the target audience, the communication situation, examples of similar tasks, definitions, and data.

In the example, I have titled this section "A/B TEST DATA". Here, I have added the test hypothesis and test data needed for ChatGPT to analyse the test. You can also upload data files, such as screenshots of the user interface you are testing, survey data, customer service data, training data like example reports your AI can learn from.

Data is a crucial part of the prompt and helps your AI understand the task it must solve.

4. Ask Your AI to Help Improve Your Prompt

Your AI is a prompt expert, so you can benefit from having it help you improve your prompt to get even better responses. Top models like ChatGPT-4, Gemini Advanced, and Claude Opus Pro are good at understanding our intentions. When I need to develop a new prompt, I always instruct my AI to ask me clarifying wh-questions like a trained journalist before it performs the task. I then answer the questions, and we have a dialogue back and forth while testing the prompt until I am satisfied.

I typically then ask my AI to summarise what we have agreed upon and add it to the prompt. This usually works, but sometimes the AI makes the prompt worse, so it’s always important to test your prompt thoroughly.

This iterative prompt development helps ensure that your AI has all the necessary information and can thus provide you with a higher quality response.

In the example, I chose to work further on the task description itself, which ended up looking like this:

YOUR JOB
Your job is to analyze the result of the A/B test described in the A/B TEST DATA section below. Summarize your analysis starting with the conclusion; using this sentence: “The hypothesis is [confirmed or rejected].” Communicate the result in an easy-to-understand language for laymen. Explain whether the result has a positive, negative or neutral impact on the [test goal]. Mention whether the winning version is [insert the confidence level percentage here but don't mention confidence level] statistically significant not, for example: "The result is statistically significant at 95%". Avoid showing how you did the calculations and avoid complex expert terms like confidence level and p-value etc.

In this way, I get a brief and precise analysis focused on the most important elements:

AB test analyseprompt

Tips for Using the Prompt Card

  • Print the prompt card, carefully consider the purpose of the prompt before filling out the prompt card by hand. This provides a better overview compared to writing directly in your AI’s small input field. It also helps you remember the different prompt elements.

  • Write clear, precise, and easy-to-understand prompts.

  • Continuously test and refine your prompts based on your AI’s output, as good prompts require multiple iterations.

  • Many prompts are interdisciplinary, so it’s a good idea to fill out the prompt card together with your colleagues to ensure that you cover the task your AI should solve comprehensively.

  • Create a prompt library for your best prompts that everyone can easily access.

  • Using a prompt card helps ensure consistent prompts across teams and projects.

Download my prompt card for free, print it out, and fill it in by hand

Share this post

Picture of Jakob Styrup Brodersen

Jakob Styrup Brodersen

I have worked with data-driven online optimization for 20 years in 5 different industries. Now, I am a freelance CRO and AI consultant: I teach and advice how to utilize the benifits of AI and I do prompt engineering.