Building Custom GPTs
Embedded Files
Learn to create Custom GPTs tailored to specific tasks, including key requirements and best practices.
Why build a Custom GPT instead of using a core model?
While GPT-4o is a powerful and versatile language model, creating a custom GPT can offer tailored benefits that better suit your specific needs. Here’s why you might choose to build a custom GPT.
Repeated Use Cases: If you find yourself using the same prompt over and over, a custom GPT can automate these tasks, saving you time and effort.
Team Benefits: When multiple colleagues can benefit from the same tool, a custom GPT ensures consistency and efficiency across your entire team.
Enhanced Functionality: Need more than what standard ChatGPT offers? Custom GPTs can integrate additional information or third-party services to provide more comprehensive and specialized answers.
Memory and Consistency with GPTs
One important aspect to consider is memory. Do you or your team need the same output every time you interact with ChatGPT? By default, each new conversation with ChatGPT starts fresh—it doesn’t remember your previous interactions.
Why This Matters
Consistency: Without memory, you might need to repeat instructions or context each time you start a new chat.
Efficiency: Starting from scratch can slow down your workflow, especially for recurring tasks.
Benefits of Using GPTs
GPTs retain context and outcomes to keep your interactions focused, reliable, and efficient—saving you time and improving your ChatGPT experience. You get the following benefits.
Reliable Outputs: Get the same quality and type of responses every time.
Streamlined Workflow: Save time by not having to re-establish context in every new conversation.
Focused Assistance: GPTs concentrate on your specific needs, making interactions more effective and satisfying.
If you have a prompt or use case you keep revisiting or something you think others on your team could benefit from, it might be time to build a GPT.
Building GPTs
1. Identify strong use cases
The first step in building a GPT is to identify strong use cases. Focus on routine activities that consistently achieve outcomes or could benefit from improved efficiency. Ask yourself:
Where can you save time on recurring tasks?
Which tasks could be streamlined?
Targeting these areas will help you create GPTs that simplify workflows and boost productivity.
GPTs for Roles
Some GPTs are useful across roles, including:
Email Drafting: Consistent formatting and tone.
Meeting Summaries: Easy-to-reference notes.
Document Reviews: Identify key points or gaps.
Department Handbooks: Comprehensive documentation.
Onboarding Guides: Effective team onboarding.
Combining these general-use GPTs with role-specific ones—like Sales Role-Play, Competitor Research, or Email Generation—can significantly boost productivity and addresse unique team needs.
Use an Impact-Effort Matrix
Use the Impact-Effort matrix (below) to prioritize your GPT use cases. Start with "Quick Wins" and gradually add complexity. Remember, GPTs excel at tasks and workflows you perform repeatedly.
Pro-tip: Save frequently used prompts as GPTs.
2. Create GPT
Clearly define your GPT’s objective to guide its development and maintain focus.
Example objectives:
Quickly answer common IT questions.
Provide step-by-step troubleshooting guidance.
Generate content matching your company's tone.
Assist users with software installation and setup.
Defining clear objectives ensures your GPT stays focused, relevant, and effective, delivering precise and satisfying interactions.
Pro-Tip: Leverage ChatGPT to help you draft your instructions. Use “Create” as a starting point to draft your GPT by telling ChatGPT your objectives. It's important to then always then switch to the “Configure” tab to manage the GPT's build & finalize it's setup.
3. Configure GPT
Toggle to the "Configure" tab and fill in the required fields.
Name
Give your GPT a descriptive name that makes it easy to find & understand its purpose.Name
Description
Add more detail on the capabilities of your GPT. This will show to users of your GPT, and is helpful to help users understand the GPT’s capabilities. Description
Instructions
Provide detailed instructions or guidelines on how the GPT should behave, its functionalities, and any particular behaviors to avoid. Use best practices here.Instructions
Conversation starters
Conversation Starters are examples of prompts users can leverage when using the GPT. They’ll show when users open the GPT as a way to start the conversationConversation starters
Knowledge
You can expand your GPTs knowledge by uploading specific files that contain proprietary information or data the model was not previously trained on. Knowledge gives your GPT the business context it needs to provide reliable answers! Gather all necessary documents, policies, procedures, FAQs, and other relevant information to attach to your GPT.See guidance on file formats, limits, and more under “Tips & Tricks”.Knowledge
Capabilities (not shown):
Web Browsing: search the web
DALL-E Image Generation: generate images
Code Interpreter & Data Analysis: Code Interpreter is a post-trained GPT-4 model that allows ChatGPT to work well with structured data. Whenever a GPT might be used with structured data (e.g. excel, csv), make sure to have Code Interpreter turned on.
Actions and Integrations
Actions enable your GPT to connect and integrate with third-party systems, allowing the model to interact with the real world on your behalf. By leveraging your scopes and permissions, Actions can perform tasks such as retrieving data, updating records, or triggering workflows seamlessly.
Benefits of Using Actions:
Extend GPT functionality with familiar tools.
Automate repetitive tasks to save time.
Enable real-world interactions beyond text generation.
Check the “Tips & Tricks” section below for guidance on building Actions.
Important: Confirm organizational policies and permissions before integrating Actions. Contact your ChatGPT admin for support.
A walkthrough of advanced GPT capabilities, including instructions, file uploads, web browsing, image generation, and code interpretation.
4. Evals: assessing your GPT's performance
Evals provide a structured way to evaluate your GPT’s performance based on your specific criteria before you start using or sharing it. Here’s how you can set up effective evaluations:
Create Evaluation Questions: Develop a set of 10-15 questions that reflect the tasks you expect your GPT to handle. Include the expected answers for each question.
Test Your GPT: Run your GPT against these evaluation questions to see if it provides accurate and reliable responses.
Analyze the Results: If your GPT doesn’t meet the expected outcomes, tweak its instructions or update its knowledge base to improve performance.
Use a Template: Utilize the provided template below, based on OpenAI’s recommended approach, to organize and create your Evals effectively.
Make sure your Evals cover a wide range of scenarios to thoroughly test your GPT’s outputs. This helps identify any potential issues and ensures your GPT can handle diverse tasks effectively.
5. Publish and share
Once your GPT is built and tested, it’s time to share it with colleagues who can benefit from it. Sharing your GPT helps streamline workflows and enhances productivity across your team or organization.
How to Share Your GPT
Share with Specific Individuals: Select and share your GPT with particular team members who need it.
Share with User Groups: Distribute your GPT to entire groups, such as the sales team, to ensure everyone has access to the tools they need.
Share Across the Company Workspace: Make your GPT available to everyone in your organization, fostering collaboration and consistency.
Steps to Manage Sharing Options
Use the Share Button: While editing your GPT, click the Share button to choose who can access it.
Access the Dropdown Menu: Alternatively, use the dropdown menu during the editing process to manage your sharing settings.
Keep in mind the knowledge and information provided in your GPT when deciding who to share with. If there is sensitive information included, use caution & only share with team members who should be able to access that information.
Tips and Tricks
Thought starters
Thought starters
Test your use case with a standard ChatGPT conversation to see if it meets your needs
For multi-phase workflows, consider breaking them into multiple GPTs
Tip: you can call multiple GPTs (that have been used or pinned) within a single conversations using ‘@’ to string these together for complex workflows
Instruction writing
Instruction writing
After creating a GPT, always update its instructions in the Configure tab. Updating in the Create tab will overwrite the original instructions.
Be clear and specific in your instructions. Break down workflows into step-by-step chunks, and introduce pauses to verify that the GPT’s output aligns with your expectations (or EVALs, as mentioned earlier).
Avoid ambiguous, negative, or weak language when aiming for specific outcomes.
Use relevant keywords or phrases from your knowledge base or third-party integrations.
Utilize punctuation and Markdown, to highlight steps or key information.
Clearly label knowledge files (e.g., “gptcorp_sales.pdf”) when uploading so the GPT knows when to use them.
Refer to our Custom Instructions Page Help Center or our Write Better Prompts page for more tips on writing the great instructions.
Iterate to improve
Refresh Configuration: Regularly update any uploaded knowledge to ensure the GPT's information is current and accurate.
Establish Feedback Loop: Create a mechanism for users to report issues or suggest improvements.
Adjust Instructions: Continuously refine instructions based on feedback and usage to enhance the GPT's performance.
Formats
Formats
Text formats: use text, PDF (text only), Word, or markdown
PDFs with graphs and table should be converted to xml (you can use ChatGPT to do this)
Numerical data: CSV or unformatted Excel files
Limits
Limits
Instructions have up to 8k characters and responses from ChatGPT can be up to 8k characters as well.
Knowledge can be up to 20 files with a max file size of 512MB per GPT.
All files are capped at 2M tokens per file (~10MB)
If you're a developer or you have familiarity with APIs, please visit our Developers at Work section to aid and accelerate you in unlocking additional value with your GPTs via Custom Actions.
Page updated
Report abuse