I will answer these questions in the following article:
- What are custom GPTs?
- What types of custom GPTs are there?
- Who can create custom GPTs?
- When are custom GPTs useful?
- How do I create a custom GPT? A step-by-step guide?
- How do I track a custom GPT?
- What is the data protection for Custom GPTs?
- Conclusion: What’s next?
What are custom GPTs?
To recap, here are a few basics: GPT assistants, developed by OpenAI, stand for “Generative Pre-trained Transformer”. This term stands for an artificial intelligence (AI) that is trained to understand and simulate human speech. They learn from a huge amount of text data to recognize patterns and contexts that enable them to respond to a variety of queries.
In addition to the well-known chatbot ChatGPT, OpenAI has also been offering the option of custom GPTs for several months. Custom GPTs is a new product from OpenAI that allows you to create your own ChatGPTs, so to speak. You will still use ChatGPT’s AI technology, but you can define how the chatbot should behave and which knowledge sources it should use. In the article Why create your own GPT assistants, I explain what types of GPT assistants there are and what you can use them for.
What types of custom GPTs are there?
If you create a custom GPT, you have 3 options for using or publishing it later.
1. private
You can simply save the custom GPT and receive an individual link that only you can use to access the custom GPT. This option is particularly useful if you want to use the GPT as your personal assistant. For example, I have created a GPT that sends e-mails for me and no one else should use it except me. The following video shows how my GPT sends e-mails for me.
2. semi-private: only with link approval
The second option is primarily intended for teams. Others can also use your custom GPT here, but only if they receive the appropriate link from you. This form is particularly suitable for custom GPTs that are created for teams
3. public
Public GPTs are, as the name suggests, public. These GPTs are displayed in the official GPT Store from OpenAI and can also be found using the search function. This option is particularly useful if you are creating a GPT that is to be used by everyone. Or if you are even pursuing the goal of earning money with your custom GPT. You can read more about earning money with GPTs in my last post.
In the following video you can see my custom GPT on the subject of Ticino.
Who can create custom GPTs?
Custom GPT can be created by all ChatGPT users who have a Plus account. This also includes users with Team or Enterprise licenses.
The same requirements also apply to the use of custom GPTs. Only those who have at least one Plus account can use their own and public Custom GPT.
When are custom GPTs useful?
Custom GPTs are particularly suitable for recurring tasks. If you work through the same tasks or workflows over and over again and want to take advantage of individual or specific support from ChatGPT, a custom GPT is a great support for you. You then no longer have to refine your own prompt each time, but can simply access the custom GPT created once.
Here are a few examples of applications where a custom GPT is useful.
- Special content bots for creating similar blog posts
- Bot for e-mail processing
- Automated analysis of data
- Tailor-made learning content
- Special content bot for social media
- Help with solving problems
- Creation of presentations and concepts
However, a custom GPT also makes sense if you want to earn money with GPTs in the long term. In my article on “How to make money with the GPT Store”, I show you some examples of how you can increase your sales with GPTs.
How do I create a custom GPT?
A step-by-step guide
1. define concept and objectives
You don’t need any programming knowledge to develop a custom GPT.
It is much more important that you have a clear goal in mind and know why you want to develop your GPT. So ask yourself these questions at the beginning:
- What problems should the Custom GPT help with?
- Who am I doing the GPT for?
- What tasks should the GPT support me and other users with?
- What must the GPT be able to do?
- What additional knowledge do I need to give the GPT?
- Do I also want to earn money with the GPT?
- Do I want to make the GPT publicly accessible?
2. start with the implementation
As soon as you have answered the majority of the questions above, it’s time for the actual implementation.
To do this, log in to ChatGPT and click on“Explore” on the left-hand side. You will then find the “Create a GPT” function at the top. Alternatively, you can click on this link and you should land directly at the right place in ChatGPT.
You will now be taken to the GPT Builder. Here you can choose between “Create” and “Configure” mode. I will now first explain how to create your custom GPT using the “Configure” mode.
3. configure the Custom GPT
First click on “Configure” and give your GPT a name and a description. These two fields have no direct effect on how your custom GPT will behave later and you can even change them again.
The same applies to the image that you can add as an icon to your own bot. You can either upload an image yourself or click on “Use DALL-E” and let the OpenAI AI create an image for you. You can repeat this process until you like the picture.
Once you have added this initial information, you can already see the name, icon and description of your GPT in the preview on the right. It is now also possible to save the GPT at the top right.
Instructions: Examples for the frame of your GPT
Now come the other details of your GPT, which will ultimately determine how your AI Assistant behaves.
In the Instructions field, you give the GPT instructions on how it should behave later. This field is therefore the most important field and will determine how well your custom GPT ultimately achieves your goals and can really support you as an assistant.
It is best to divide the instructions into the areas of what he should definitely do and what he should not do under any circumstances. Here are some examples of instructions:
- Behave like an experienced copywriter who writes blog articles in a professional but not too uptight style. You always address your readers directly and write from a “we” perspective. The sender is Sophie Hundertmark’s company.
- You always ask about the topic first.
- You also create the outline. Then ask questions to refine them further.
- You integrate images in appropriate places.
- Always structure article outlines in a similar way. First there should be an explanation of the topic, then the main part, and at the end a FAQ section.
- Avoid repetitive formulations and write in a varied way.
My tip: Write as many detailed instructions as possible. Sometimes it even makes sense to repeat important instructions several times in different words.
Conversion starters: examples of conversation starters
ChatGPT refers to the small suggestions or boxes that are placed above the input field of the GPT as conversion starters. Here you can, for example, place recurring tasks or sample questions that others can ask your GPT.
Knowledge: Upload your own files for training
Every artificial intelligence needs to be trained in order to function. ChatGPT has of course already integrated a very broad knowledge base, which will initially benefit your custom GPT.
Nevertheless, you still have the option of uploading further content to your GPT at this point. This way you can make your GPT even more specific.
You can even say in the instructions that your GPT should prefer these knowledge sources. In our Greenwatch Assistant, for example, we have added a large number of regulations and draft laws on greenwashing as primary sources of knowledge.
Capabilities
One of the announced functions of the ChatGPT alternative Google Gemini is multimodality. This means that I can no longer just output text or images, but both are combined in one AI – or even more, the two options (image and text) are just examples. This is also possible with custom GPTs. Here you can select which functions your custom GPT should contain. You can choose between Web Browsing, DALL-E (image generation) and the Code Interpreter.
Actions
At this point you will need programming skills for the first time if you want to develop your custom GPT further. Here you can link the GPT with interfaces and thus give it even more functions. My personal GPT for sending e-mails uses the interface to make.com and can therefore access my e-mails.
4. test the GPT
Now it’s time to test your GPT in the preview. You can use the preview field and ask your GPT questions.
As with other chatbots, this test phase is extremely important to ensure that the GPT really fits your goals and your target group. So take your time and ask the GPT enough questions or give it appropriate commands. You can continue to adjust the GPT in the left-hand field parallel to the test.
5. saving and publishing the GPT
When you have completed your tests and are satisfied, click on “Publish” in the top right-hand corner. You then have the following options:
- Only me: If the bot is only to be used by you.
- Only people with a link: Similar to a hidden YouTube video, the bot is only accessible via a link. This method is best if your GPT is intended for the whole team.
- Public: The GPT appears in the GPT Store, which is accessible to everyone worldwide.
6. customize GPTs
Of course, you can adapt, change and further develop your GPT at any time. You can go back to Configure mode at any time and adjust the individual fields.
Chat (Create): Create the GPT with ChatGPT
As described above, you can choose between Create and Configure at the beginning. If you select “Create”, ChatGPT will guide you through the dialog for creating the GPT. This can be helpful if you don’t yet know exactly what you want your GPT to be able to do. With this option, ChatGPT asks you a variety of questions about what your GPT should look like in the end and what behavior it should have.
However, my experience has shown that this method of implementation usually leads to rather inaccurate results.
How do I track a GPT?
We are used to tracking our tools and activities so that we can continue to develop them in a meaningful way. However, there are currently no options for analyzing a GPT. You cannot see which questions have been asked and how other users have interacted with the GPT.
What is the data protection for Custom GPTs?
To date, there is no detailed information on data protection or the rights of Custom GPTs. Basically, the same “rules” apply here as with ChatGPT. It can be assumed that the information you enter in a custom GPT can also be viewed by OpenAI.
Conclusion: What happens next?
It’s best to test the possibilities of custom GPTs for yourself. And if you would like further support or perhaps you are missing the right use case at the beginning, I will be happy to help you as always. Write me a non-binding e-mail or WhatsApp message and we’ll have a chat.
I would also like to invite you to test my own GPT Assistant in the GPT Store.
