How to create a Custom GPT on ChatGPT? Complete guide + demo

Since November 2022, ChatGPT has revolutionized the way we interact with the web. Many companies have succeeded in increasing their productivity by changing the way they work. That's why it's essential not to miss the boat on getting into the habit of using these new "Gen AI" technologies.

OpenAI's latest innovation is the creation of GPTs. In other words, the ability to create a ChatGPT totally customized to your needs. We've put together a complete guide and tutorial to help you create your own GPTs.

What are GPTs?

GPTs (Generative Pre-trained Transformers) are customizable versions of OpenAI's ChatGPT, designed to function as "agents" that can be programmed to perform specific roles.

‍

How to use GPTs

Your GPT can be configured to browse the Web, generate images using DALLE-3 or execute code.

‍

Since January, users with a ChatGPT Plus or Team account can already personalize ChatGPT responses with customized instructions, in 2 different ways: 

• Create multiple custom GPTs: There's no limit to the number of GPTs you can create, making it easy to switch between chatbots and get the right kind of AI-based assistance for the occasion. Customized instructions, on the other hand, limit you to one set of instructions per user. 

• Adding files as sources. Instead of copying and pasting the text of every resource you want ChatGPT to take into account when generating a response, you can upload knowledge files directly into your GPT builder.
  ‍

It's important to note that OpenAI won't be able to access anything that contains sensitive information, such as your chat history and knowledge files, to train its models. But as with any generative AI, there's always the risk of data breaches and ways to access your knowledge sources, so be careful what you download in the early days of this feature.
‍

How to build your own customized ChatGPT

We'll show you a short method for customizing your chatGPT using OpenAI's GPT builder with the following steps.

1. Login to OpenAI

You need a Plus or Team subscription to access GPT-4. Start by logging on to your https://chat.openai.com/ account. 

2. Create your GPT

1. Once on the ChatGPT interface, on the left-hand menu, go to the "Explore GPTs" section.

2. Once in the section dedicated to GPTs, click on "Create" and the creation interface appears. The screen is divided into two parts: the Create panel on the left is where you enter your prompts to build your chatbot; the Preview panel on the right lets you interact with your chatbot as you build it, making it easier to determine how to refine it.

3. You can now start creating your GPTs. Enter your prompt in the "Message GPT Builder" field on the Create page.
4. The GPT builder will then suggest a few elements based on your instructions: a chatbot name, a profile photo, ... You can accept the suggestions or ask ChatGPT to modify them. If you accept the initial suggestions, you can always modify them later.
   ‍

The GPT builder will prompt you to enter more specific instructions to fine-tune your chatbot's behavior.
‍

If you're not sure how you want to modify your chatbot's behavior, that's okay. I suggest testing your chatbot in the preview panel, interacting with it as you normally would, and using its responses to refine your prompt. For example, if your chatbot generates long responses, you can ask the builder to make the responses short.

5. Keep refining your requests until your chatbot starts returning the type of answers you want your final chatbot to generate. You can ask it not to quote your competitors and list their names, for example, or tell it that if it doesn't have the answer, the user is invited to send an e-mail to a specific address.

3. Configuring your GPT

Now that you know how to create your custom GPT, you can further customize it with advanced settings via the "Configure" tab.
‍

In the "Instructions" text field, you can enter further instructions on how your chatbot should or shouldn't behave.

The "Conversation starters" setting lets you add conversation starters that chatbot users can choose from, to make it easier for them to find answers to their questions.

For the "Knowledge" part, you can provide OpenAI with one or more relevant files to which your personalized chatbot can refer and provide answers.

You can also ask your chatbot to retrieve external information using APIs, thanks to the "Actions" function.

4. Save your GPT

‍

‍

When you're happy with your chatbot, click "Save" to save it. You can then publish your custom GPT and choose who you'd like to share it with, or whether you'd like to keep it private.

Going further: publish your custom GPT on the OpenAI store

1. Having the right plan

To publish your custom GPT on the OpenAI store, you need a Plus, Team or Enterprise subscription, as the free version does not allow this.

2. Check your profile information

In order to publish your GPT for all to see, you need to go to your account's "Settings" and click on "Settings & Beta", then go to the "Builder Profile" section. Choose your name to publish GPTs under your name. To publish under a website, click on the "Select a domain" drop-down list and enter your domain name.

3. Publish your GPTs
   ‍

You'll need to go back to your custom GPT and modify it. Click on the "Save" button and select "everyone" so that your chatbot is visible to everyone, then click on "Confirm". OpenAI will check your GPT, and once it's validated, you'll be able to find it in the store.

Demo: Creating your own chatbot with OpenAI

On February 8, 2024, Hugo Cornu hosts a live and presents how to create a chatbot based on your data. The video is available right here : 

Conclusion

To create an effective custom GPT, it's essential to clearly define your expectations. By following the steps outlined in this article, you can quickly create your first custom GPT and test it in a private environment to get started. If you want to make your GPTs visible on the stores, you'll need to upgrade your account. In any case, it's important to keep in mind that for ChatGPT's chatbot to provide relevant answers, it needs to be regularly updated and improved.

‍