I’m always on the lookout for ways to make life easier for myself and my clients. Recently, as I was using Chat GPT to summarize the transcript of my monthly MatrixMaxx AMS Q&A session, I realized that the process I use every time was the same. I could probably create a custom GPT to both make the process faster and also have the results be more consistent from month to month. And let me tell you—it’s been a game-changer. Hours of time saved, times 12 months a year, means a huge windfall of time for me to do more meaningful product work!
This awesomeness didn’t happen overnight, though, so let me walk you through the process and share some key insights I picked up along the way.
What is a Custom GPT?
Custom GPTs are a no-code feature of ChatGPT, available to all paid accounts of ChatGPT, that lets users customize the chatbot according to the specific way they use it, instead of having to give full instructions everytime they engage in a chat.
Creating your own GPT allows you to enter all the instructions once and save it. You also do not need to be as familiar with prompt engineering techniques. You give the instructions in plain text, adding additional needs as you fine-tune the GPT and the builder produces the instruction prompt. It also allows you to enter sample documents to match tone and formatting.
There are additional benefits of making your own GPT that I didn’t use in this case, but are good to know about, like using the knowledge settings of the GPT Builder. You can upload specialized knowledge like reports or other documentation that the GPT should pull from first, before going to the rest of the Large Language Model (LLM). This works like a personal library that you can query against. You can also add an API connection to an external database for even more knowledge to be included.
A note about your data: ChatGPT offers multiple levels of accounts. Some of these will take any data you enter and add it to its LLM for future learning. The higher levels do not take your data and add it to the LLM, which is important if you are working with proprietary data.
At the time of writing this, the Free, and Plus account do add your data to the LLM. The Team, Enterprise, and API accounts do not. I use a Team account and I can share any GPTs I create with the other members of my team.
Step-by-Step of Creating My Custom GPT
Let’s walk through the nuts-and-bolts of how I created my GPT.
- To start, I went to the “My GPTs” section in ChatGPT. I then clicked to “Create a GPT.”
- The GPT builder will come up on the Create view and look something like this:
- Start with the basic overview of what you want this GPT to do. You don’t need all the details at this point. You can add details and further instructions as you go.
- In my case, I started with:
“This GPT is a Customer Success Manager specializing in web software, assisting users in summarizing and formatting transcripts from Q&A sessions. Summaries include main topics that were covered along with timestamps that each section stated.”
- The builder will then ask you to name the GPT.
- Next, it will suggest an image for your GPT. You can either give instructions for it to create a different image, or you can upload your own.
- It will then ask a series of questions like “what should be emphasized or avoided?” But at any time you can start to give your own instructions.
- I copy and pasted previous summaries and told it to use them as examples. That added tone guidance to the GPT instructions.
- I even added specific formatting instructions like “include the starting timestamp immediately after the main topic areas” and “make the main topics formatted as <h2> with the details bulleted underneath.”
- Finally, I gave it the instruction to always correct “matrix max” to “MatrixMaxx” as that is something the transcript always gets wrong.
At this point, you can go ahead and start using your custom GPT by engaging with the chat on the right. In my case, all I would need to do is paste the transcript into the chat. But, if you want to get more granular, you can click on the Configure tab at the top. For my GPT, it looks like this:
You can see this is a more direct place to edit your GPT’s instructions. You can also see how it has distilled all the plain language instructions I gave it into one set of GPT instructions. The instruction for my GPT now says:
“This GPT is a Customer Success Manager specializing in web software, assisting users in summarizing and formatting transcripts from Q&A sessions. It emphasizes the clarity of the topics covered, ensuring users can easily identify each major topic. Summaries include clear timestamps placed immediately after each topic title, allowing users to navigate to key areas efficiently. The summaries follow the format where each topic title is formatted as an H2 header, and detailed descriptions of key points are provided in the body as bullet points. Timestamps are still placed immediately after each topic title. The GPT focuses on condensing key points, enhancing clarity, and organizing information in an accessible format. The length of detail matches a well-rounded but concise example format, as illustrated by the given session example. The GPT balances brevity and thoroughness, ensuring users receive concise but complete summaries. It provides suggestions on format and structure for better readability, while remaining professional and helpful in tone. When additional details are missing or unclear, it fills in the gaps without overstepping user intent, asking clarifying questions only when necessary. It can handle long transcripts and extract the most valuable insights efficiently. Additionally, when the phrase ‘matrix max’ appears in a summary, the GPT automatically corrects it to ‘MatrixMaxx’ to ensure consistency and accuracy in branding.”
I can easily fine-tune things from here, or add additional instructions here or through the plain language builder on the create tab.
The Result
My custom GPT exceeded my expectations. Now, not only do I not need to create the same prompt every month, but the results are much more consistent. Giving it the extra formatting instructions also saves time to get the summaries out of ChatGPT and onto our Support Center.
This is going to allow our team to amass our own prompt library without manually retyping specific prompts. We can create multiple GPTs that provide different specific outputs based on our prompts. Creating your own GPTs also lifts the burden of prompt engineering by building the prompt for you based on your plain language instructions.
And, if you think creating your own GPT is cool, checkout the GPTs available in the store that have been created by others. You can find these under the “Explore GPTs” section if you have a paid ChatGPT account. You will find GPTs by names you recognize, like WIX and Canva, and also by many other users who have found interesting uses for GPTs.
If you find a task that you do repeatedly that you think a Large Language Model like ChatGPT could help with, check out making your own GPT.