How to Create and Sell GPTs

To create your own GPT, you’ll need a paid ChatGPT account—such as Pro, Teams, or Plus.

While free users can access and use existing GPTs, the ability to build or customize one is restricted to paid plans.

ChatGPT has indicated that it will highlight the most helpful GPTs created by users and is exploring ways to support GPT creators through future monetization options.

GPTs are intended for use within the ChatGPT platform only.

OpenAI has specified that GPTs cannot be embedded into external websites using platforms like WordPress, Squarespace, or other site builders.

Instead, they provide an API for integration purposes outside the ChatGPT interface.

Here are three final key points:

1. GPTs are seen as an extension of ChatGPT. They allow users to tailor the chatbot to perform specific tasks or follow custom instructions, making it more useful for personal or professional needs.

2. No coding is required to build a GPT. The creation process is designed to be intuitive and logic-based, so anyone can create one without technical expertise.

3. GPT development aligns with OpenAI’s broader mission. As more users customize and engage with ChatGPT, OpenAI gathers insights that contribute to the development of AGI—Artificial General Intelligence. AGI refers to the ability of computers to learn, reason, and initiate tasks independently.

Start with the Wizard

Assuming you have a Pro, Teams, or Plus account, you’ll start by navigating to the sidebar in ChatGPT, where you’ll see the “Explore GPTs” option.

Click on it to view a selection of existing GPTs for inspiration. To begin creating your own, click the “Create” button on the right-hand side.

This opens the GPT creation interface, which offers two paths:

1. Use the configuration tab to manually build and customize every aspect of your GPT.

2. Use the step-by-step wizard, which walks you through the process in plain language—ideal if you’re new or prefer a guided setup.

The wizard allows you to describe your GPT’s purpose in everyday language. No coding is required—just a clear idea of the problem your GPT will help solve.

On the right side of the screen, you’ll see a live preview of your GPT as you build it. 

You’ll begin by defining the GPT’s intended use. Don’t worry if you get it wrong—this initial prompt can always be edited later.

You can also upload files or images during this first setup phase, including importing content from Google Drive or Microsoft OneDrive to help inform the GPT’s functionality.

Once your initial prompt is entered, ChatGPT will process your input and prompt you to name your new GPT.

If you plan to share or sell it later, this is a good time to think carefully about your naming strategy, as it may influence discoverability and appeal.

Naming Your GPT

At this stage, it’s a good idea to start thinking about marketing considerations, especially when naming your GPT.

If you have relevant documents, you can upload them here to give ChatGPT additional context. These materials can help shape the GPT’s identity and direction.

Alternatively, you can engage in a naming conversation directly with the chatbot.

Ask it for suggestions, or let it ask you questions to better understand your goals.

The wizard will propose a name, but you’re free to refine it or explore other options through this back-and-forth dialogue.

As part of the setup, ChatGPT may also suggest a profile image to go along with your GPT. If you’d prefer, you can choose to upload your own image instead.

This is a great opportunity to use your favorite AI image generator to create something custom and on-brand.

The process is intentionally flexible, allowing you to iterate on the name and visual identity until you’re happy with how your GPT will appear to others.

Custom Image for GPT

To create a custom image for your GPT, it’s best to open a separate conversation within ChatGPT.

While you could use external tools like Midjourney or Canva, ChatGPT can also generate the image for you—especially if you let it know the image will be used as part of a GPT.

This context helps ChatGPT design something that visually fits your use case.

Start by switching to the GPT-4o model, enabling tools, and selecting the image generation option. Then, clearly describe what you want the image to convey.

Be sure to mention that it’s for use in a GPT, as that influences layout, tone, and perspective.

You can also upload existing branding elements—such as a logo or color reference—to help ChatGPT align the image with your visual identity.

This is why working in a separate chat is helpful: it allows you to guide the design process with more detail and incorporate branding cues.

Once the image is created, review it.

You may want to iterate a few times with ChatGPT to fine-tune the style, layout, or tone of the image before settling on a final version.

Once satisfied, download the image and return to the GPT creation interface to upload it as your GPT’s profile image.

If you’re planning to build multiple GPTs and want a consistent visual brand, you can use the same process to generate similar images that share a common look and feel.

Keeping a record of your images and design parameters makes it easier to scale your GPT creations with cohesive branding.

As you continue through the wizard, keep in mind that every interaction may differ slightly.

For example, you may now be prompted to refine how your GPT communicates or what it emphasizes—another chance to tailor it for your audience and purpose.

Refining the GPT 

At this stage, you’ll be prompted by the chatbot to refine how your GPT communicates.

Simply respond by indicating that you’d like to adjust its tone or style, then press Enter.

ChatGPT will ask an open-ended question to help guide this customization.

You’ll be asked whether the GPT should sound formal, conversational, or like a seasoned advisor, among other options.

If you have a sample—such as an interview transcript or writing example that reflects the voice you’re aiming for—you can upload it here.

Alternatively, you can describe the desired communication style directly in the chat.

You have considerable flexibility here.

Whether you want your GPT to use industry jargon, be friendly and casual, or speak with authoritative precision, this is the time to define those expectations.

It’s helpful to think about your end user:

• Who will be using the GPT?

• What kind of experience do you want them to have?

• How should the GPT interact to build trust and encourage repeat use?

Type your preferences clearly in response to the prompt, and ChatGPT will begin applying those parameters. You may see follow-up questions to clarify further.

Keep in mind, each wizard interaction may differ, so your experience may not follow this script exactly—but the intent will be similar.

After submitting your tone and style preferences, ChatGPT will update the GPT accordingly. At this point, it will notify you that your GPT is ready for a preview.

Even if the configuration isn’t complete, it’s a good idea to start testing now.

This allows you to verify whether the GPT is responding in the way you intended and to make any adjustments before finalizing.

Testing and Previewing

When your GPT enters preview mode, you’ll notice a few elements worth reviewing—particularly the conversation starters.

These are suggested prompts for users to begin interacting with your GPT. However, they may not reflect the experience or tone you’re aiming for.

If needed, you can adjust them later by going into the Configure tab.

Similarly, the description of your GPT might not fully align with your intended use. This too can be edited within the configuration settings.

At this stage, your goal is to evaluate whether your GPT is functioning as intended.

Walk through the wizard and test how the GPT guides users through its responses.

Read through the answers it generates to determine whether it’s heading in the right direction.

If the GPT is generally performing well but needs a few adjustments, now is a good time to transition into the Configuration settings for fine-tuning.

By default, your GPT will run on the latest model—typically GPT-4o.

This is often the best choice for most use cases, but depending on your GPT’s purpose, you might want to explore other models.

For example, if your GPT relies on advanced reasoning, it may benefit from a model better suited to complex problem-solving.

You can change the model and test the difference directly within the preview environment.

If your testing shows that your GPT is functioning well overall, you’re ready to begin finalizing the setup.

Continue working with the wizard, and confirm that your GPT is moving in the right direction based on your initial vision.

Once you do, click Enter to proceed.

The wizard may offer additional customization options based on the type of GPT you’re building.

If so, this is the time to explore and complete those steps before moving into full configuration.

Call to Action

You might be wondering whether it’s possible to include a call to action (CTA) within your GPT—either during the wizard setup or when refining its output.

To clarify this, we asked ChatGPT directly.

The response: Yes, you can guide users to a webpage for more information.

You’re allowed to include creator links within your GPT’s custom instructions or in the output text. However, there are important guidelines to follow:

• Link to a neutral landing page, not directly to a checkout or sales page.

• If your GPT will be made publicly available in the ChatGPT marketplace, you must clearly disclose this kind of redirection to users upfront.

• Avoid prohibited practices, such as:

  • Direct monetization or affiliate links

  • Misleading or coercive messaging

  • Lack of transparency or failure to disclose intent

That said, if you’re the GPT creator, you can include something like:

“If you’d like a free worksheet to apply this framework, visit [yourwebsite.com]. I built this to help you solve [specific problem the GPT addresses].”

This kind of value-based CTA is acceptable—as long as it aligns with ChatGPT’s usage policies and offers genuine help to the user.

Keep in mind that guidelines may evolve, so it’s wise to double-check with ChatGPT directly at the time you’re creating or publishing your GPT, especially if your goal involves directing traffic, generating leads, or offering digital resources.

Tweaking the Instructions 

At this stage, we’re inside our ChatGPT project, and we’re ready to refine the instructions for how the GPT should behave.

To do this, we’ll ask ChatGPT for feedback and guidance on improving them.

After clicking Enter, ChatGPT will analyze the instructions and determine whether they’re strong enough to support your GPT’s intended purpose.

One useful feature is that you can ask ChatGPT for a critical review of your instructions.

There’s no special phrasing required—just give it permission to be honest and thorough in its feedback.

At this point, you have two options:

• You can ask ChatGPT to completely rewrite the instructions, or

• You can work point by point, collaborating to refine each section.

Which path you choose depends on how polished your current instructions are and how specific your GPT’s goal is.

In general, it’s worth taking the time to create a thoughtful instruction set—one that ensures users have a smooth, helpful experience with your GPT.

For this example, we’ll take the quicker route and simply confirm that we’d like ChatGPT to rewrite the full instructions for us.

ChatGPT may then ask whether you’d like the output formatted specifically for pasting into the GPT setup interface. When prompted, say yes, and press Enter.

ChatGPT will then generate a clean, ready-to-use set of instructions that you can copy and paste directly into your GPT’s configuration.

Remember: every interaction with ChatGPT is tailored to your specific use case, so you may receive slightly different follow-up questions based on the details you’ve provided.

Recipe or Direct Access 

ChatGPT has the ability to export content into Word documents or PDFs, which can be especially helpful during the GPT creation process.

As you build and refine your GPT through conversations, it’s a good practice to request a compiled output of the relevant dialogue, formatted as a downloadable document.

This becomes particularly useful if you want to offer a “GPT recipe”—a documented set of instructions and configurations—rather than sharing a custom-built GPT link.

For instance, if you’re selling your GPT as a one-to-many solution, it might be more beneficial to the customer to receive a guide they can customize and upload themselves, rather than relying on direct access to your prebuilt GPT.

This distinction is important. Your delivery method depends on your business model and goals:

• If your intention is to help customers build their own GPTs, you should regularly download and compile your build conversations into shareable resources.

• If your goal is for users to interact directly with your custom GPT, you’ll still want to save your configuration and development conversations—but primarily for internal reference, not for distribution.

In short, be intentional about how you document and repurpose your GPT creation process.

Whether you’re offering a plug-and-play experience or empowering users to build their own, capturing your work in exportable formats adds professionalism and flexibility to your offering.

Welcome Message 

When we updated the model, ChatGPT provided a suggested welcome message to accompany the rest of the configuration.

However, we noticed that there wasn’t a clear place to enter the welcome message within the Configure tab.

To resolve this, we switched over to the Create tab and informed ChatGPT that we wanted to add a welcome message. ChatGPT generated one accordingly.

Now, we’re going to update that welcome message using revised content from a previous conversation.

Even though the Configure tab doesn’t visibly display a field for the welcome message, you can input or update it directly from the Create view.

So we’re now pasting the updated message into the chatbot, and ChatGPT is applying the changes.

During this step, ChatGPT flagged that the welcome message was too long for the allowed parameters, so it automatically shortened it to fit.

Meanwhile, we noticed that our profile picture disappeared, likely during the update process.

We re-uploaded the image, and now both the profile picture and the welcome message have been successfully reset.

To summarize: if your welcome message exceeds the system’s character limit, ChatGPT will help trim it to size.

And while the interface might not always make it obvious where to place this message, you can manage it easily through the Create tab.

Redoing the Conversation Starters 

We reviewed the conversation starters and noticed that the ones displayed in the Configure tab are the same ones users see when they first interact with your GPT.

These default suggestions may not align with your tone, goals, or the experience you want to offer.

Fortunately, you can customize these conversation starters to better reflect your GPT’s purpose.

A great way to do this is to open a separate ChatGPT conversation and share your intentions.

That’s exactly what we did—we took a screenshot of the current conversation starters, explained our objectives to ChatGPT, and asked for revised options.

ChatGPT responded with a new set of conversation starters—both replacements for the existing ones and a few additional suggestions to consider.

From there, we simply copied and pasted the preferred starters into the Configure tab.

Customizing your conversation starters is an easy but powerful way to improve the user experience.

By leveraging ChatGPT’s assistance in crafting these prompts, you ensure that your GPT feels engaging, relevant, and purpose-driven from the very first interaction.

Capabilities Enhancement

As a follow-up to the previous step, we successfully updated our conversation starters without needing to delete and re-enter them manually.

Instead, we went to the Create tab, pasted the new conversation starters, and asked ChatGPT to replace the existing ones.

Once processed, the updates automatically synced and appeared in the Configure tab.

As you continue working through the wizard, ChatGPT will determine which capabilities to enable by default.

In our case, features like web browsing, GPT-4o image generation, and the canvas tool were enabled.

However, code interpreter and data analysis were not included by default.

If you believe your users would benefit from features like file upload or data analysis, you can manually enable them in the Configure section.

This is a good opportunity to tailor your GPT’s functionality based on your audience’s needs.

If you’re unsure what each feature does or how it contributes to your GPT’s purpose, consider copying the capabilities list and pasting it into your ongoing ChatGPT conversation.

Ask for feedback on how to enhance the user experience using these tools.

In response, ChatGPT may not only explain the features but also suggest new conversation starters and enhancements to your instructions that align with the activated capabilities.

To apply these suggestions, simply return to the Create tab, describe the update you’d like to make, and let ChatGPT adjust the GPT accordingly.

Once the changes are applied, your configuration will be updated, and the new instructions or conversation starters will appear in the corresponding sections.

Keeping an ongoing dialogue with ChatGPT about your GPT ensures continual improvement and helps you deliver a more engaging, useful experience for your users.

Recommended Model

One of the newer features within ChatGPT is the ability to recommend a specific model for your GPT.

If the user has access to that model, they will automatically use it when interacting with your GPT, based on your recommendation.

You have full control over which model you recommend.

When a user opens your GPT, it will default to the model you’ve suggested, provided they have access.

This allows you to optimize the experience based on the GPT’s intended purpose.

However, before finalizing your choice, it’s a good idea to consult ChatGPT and ask whether a particular model is best suited for your GPT’s capabilities.

In most cases, the default (latest) model—such as GPT-4o—will be the most appropriate.

It offers broad functionality, high performance, and the widest accessibility.

That said, you also have the option to leave the model selection up to the user, which can offer maximum flexibility.

To summarize:

• Recommending GPT-4o is ideal for stability and feature access.

• Leaving it open (no recommendation) empowers users to choose the model they prefer.

• Always confirm with ChatGPT whether a specific model offers advantages for your GPT’s use case.

Knowledge Files 

Within the Configure tab of your GPT setup, you’ll find a Knowledge section. This is where you can upload files to support and enhance your GPT’s performance.

These files serve as a reference for your GPT and help shape its responses—ultimately improving the experience for your users.

However, there are a few key things to keep in mind:

✅ File Upload Limits:

• You can upload up to 20 files per GPT.

• Each file must be under 20MB.

• Supported formats include: .pdf, .txt, and .csv.

To ensure your GPT runs efficiently and accurately:

• Avoid redundancy. Your GPT pulls information from both the system instructions and uploaded files. Repeating content can confuse the model or dilute your intended outcomes.

• Use files to reinforce the behavior and tone you want your GPT to reflect—especially how it interacts with users.

⚠️ What to Avoid:

• Scraped content from websites or FAQs without permission.

• Copyrighted materials or internal documents you don’t own.

• Poorly formatted scans or documents with unclear structure.

• AI-generated filler text that doesn’t directly support your GPT’s purpose.

According to ChatGPT, when it comes to training materials, less is more.

A concise, focused one-page text file can be more effective than a lengthy, unfocused PDF. Aim for clarity, specificity, and purpose in each document.

💡 Branding and Differentiation:

For creators offering their GPT publicly or commercially, consider uploading original, proprietary content that reflects:

• Your unique voice or teaching style

• Your brand identity

• Specific logic, frameworks, or analogies that represent how you explain and solve problems—not just how ChatGPT might by default.

Well-curated files can strengthen your GPT’s USP (Unique Selling Proposition) and help distinguish it from generic offerings.

They also support consistent behavior, messaging, and tone—especially if your GPT is part of a branded product or business solution.

Use ChatGPT to Review the Uploads

If you’re working on your GPT project and actively chatting with ChatGPT about how to structure it, you can also ask for feedback on the documents you’re planning to upload to the Knowledge section.

Simply upload your files within that same conversation, or—if you prefer—you can open a separate thread dedicated to evaluating these materials.

In our case, we asked ChatGPT to review our documents and assess whether they would be valuable additions to the GPT’s Knowledge base.

When doing this, it’s helpful to invite honest, critical feedback—you can even assign ChatGPT a specific role (e.g., content editor, brand voice consultant), though it’s not required.

The key is to be clear about your goal: to determine whether these documents are relevant, effective, and aligned with the intended use of your GPT.

One of the key benefits of this process is that, even if a document isn’t ideal as-is, you’re still working with original content.

This means you can ask ChatGPT to revise or rewrite the materials to better reflect your goals, enhance your brand voice, and fit within the file size and formatting constraints.

After reviewing and refining, you can download the revised version and then upload it to your GPT via the Knowledge section.

In short, ChatGPT can help you:

• Evaluate whether a file will meaningfully support your GPT

• Enhance content to better reflect your tone, brand, and goals

• Repurpose rough drafts or informal notes into polished training materials

Ellipsis Controls 

Within the Configure tab of your GPT, you’ll notice a set of ellipsis (⋯) controls at the top of the screen.

Clicking on these provides you with several important management options.

Key Functions of the Ellipsis Menu:

• Copy Link:
  You can copy a shareable link to your GPT—but this option is only available once your GPT has been published. If your GPT is still in draft mode, the link option will not appear. You’ll need to publish and update your GPT before sharing it publicly.

• Version History:
  This feature allows you to view and compare past versions of your GPT. You can see exactly what changes were made over time, compare versions side by side, and revert or reference older iterations as needed. You can exit comparison mode anytime.

• Duplicate GPT:
  You can make a copy of your GPT, which is useful if you want to experiment with a different version or tailor it for another use case—without altering your original.

• Delete GPT:
  If you no longer want to keep a particular GPT, you can permanently delete it from the ellipsis menu. Once deleted, that GPT will no longer be accessible.

These additional controls offer flexibility and version control, making it easier to maintain one version of your GPT for personal use while sharing or customizing another for public or client-facing applications.

Final Check 

At this stage, we want to reverse the process and ask ChatGPT if there’s anything else we should add to improve our GPT. To do this effectively, we’ll:

1. Retrieve our current instructions

2. Capture a screenshot of all visible configuration elements (if possible)

3. Upload any supporting documents we’ve included in the Knowledge section

We’ll then provide all of this to ChatGPT and ask for suggestions on how to further enhance the GPT.

ChatGPT will typically respond with thoughtful recommendations—such as additional functionality, instructional refinements, or ways to improve the user experience.

In many cases, it may even suggest missing features or enhancements based on what it sees.

We’re choosing to accept and apply all of ChatGPT’s suggestions.

This process works especially well because we’ve been maintaining an ongoing, dedicated conversation about this specific GPT project in ChatGPT.

Keeping a separate conversation thread focused solely on the GPT build helps ensure continuity and consistency.

Next, we’ll ask ChatGPT to format its suggestions in a way that can be copied and pasted directly into the Create tab.

Once we receive that formatted content, we’ll return to the Create tab and ask ChatGPT to integrate the updated instructions or elements.

Before finalizing, we’ll also re-upload our profile image to ensure everything reflects our latest changes.

Once complete, we’ll update the GPT with the new content and visuals in place.

Share Link 

With the setup process complete, we’re now ready to share our GPT.

Our goal is to provide controlled access—allowing customers to use the GPT while keeping it private from the general public.

To do this:

1. Click the “Share” button at the top of your GPT workspace.
   Up until now, access has been set to “Invite Only,” meaning only designated users could view or interact with the GPT.

2. Click the dropdown menu and change the access setting to “Anyone with the link.”
   This allows access for those you specifically share the link with—without making the GPT publicly available in the GPT store.

3. Important:
   Do not select the option to list the GPT in the public store unless your intention is to make it fully open to all ChatGPT users.

4. Next, confirm that “Anyone with the link can chat with this GPT” is selected. This ensures users can interact with the GPT but not modify it.

5. Once these settings are applied, click “Save.”

6. Finally, open the ellipsis (⋯) menu, choose “Copy Link,” and paste the link into an incognito browser window to test access and ensure everything is functioning as intended.

Sales Page

To complete the process, we’re now returning to the original ChatGPT conversation we used to create our GPT.

This is important because ChatGPT already has the full context from that thread, including your goals, features, tone, and branding.

We’ll now prompt ChatGPT to write a sales letter based on everything it knows about the GPT.

If you’re familiar with specific copywriting frameworks (such as AIDA, PAS, or Problem–Agitate–Solve), it’s helpful to mention that in your request to guide the structure and tone of the copy.

Once the sales letter is complete, you can ask ChatGPT to generate the output as an HTML file—making it easy to use as a standalone web page or landing page.

You’ll then receive a downloadable HTML file ready for customization.

At this point, you can:

• Insert your own sales links or calls to action

• Host the HTML page on your website or sales platform

• Use the content to market your GPT to a specific audience

This process makes full use of the GPT creation conversation history, allowing you to repurpose everything you’ve built into a compelling offer for customers.