Custom GPTs are your own purpose-built versions of ChatGPT. You give them a name, a set of instructions, and optionally a knowledge source or external connection — and they deliver consistent, on-point results every time, without you re-explaining the context.
A Custom GPT is a version of ChatGPT you configure yourself. It has your instructions baked in, your knowledge sources connected, and your preferred conversational style set — so every conversation starts exactly where it needs to.The GPT Builder is available to all ChatGPT Plus, Team, and Enterprise users. You do not need to write code.
Regular ChatGPT
Custom GPT
You re-explain context every time
Instructions are persistent
General-purpose
Built for a specific task
Personal use
Shareable across your team
No memory between sessions
Can reference your documents
Enterprise stat: According to OpenAI’s March 2025 enterprise report, users who work with Custom GPTs save an average of 40–60 minutes per day on repetitive tasks. 20% of all enterprise messages are now processed via a Custom GPT or Project.
In ChatGPT, click your profile icon in the top right, then select My GPTs. From here you can see GPTs you have created or have access to, and click Create a GPT to open the builder.
The Configure tab
The builder has two tabs: Create (a conversational setup flow) and Configure (manual fields). Always use Configure — it gives you full control over every setting and is faster once you know what you are doing.
Before you open the builder, define these three things. GPTs without clear planning are hard to test and easy to misuse.
1
One-sentence task description
Write exactly what this GPT does. Example: “This GPT helps account managers write personalised follow-up emails after client meetings, using the notes the user provides.”If it takes more than one sentence, narrow the scope.
2
What it needs to know
List the documents, policies, or templates your GPT should reference. These become your knowledge sources. Supported formats: PDF, Word (.docx), plain text. In 2026, you can also connect Google Drive or Slack natively.
3
Who will use it
If it is just for you, keep it private. If your team will use it, decide on sharing before you build — ChatGPT Team workspaces let you publish GPTs internally so the whole organisation can find and use them.
The Instructions field is the most important configuration you will make. This is the persistent system prompt — it tells the GPT what to do, how to respond, what to avoid, and what format to use.Example instruction structure:
You are a follow-up email writer for the account management team at [Company]. When the user shares meeting notes, write a personalised follow-up email that:- Opens with a specific reference to something discussed- Summarises the key agreed actions (numbered list)- Closes with a clear next step and proposed dateKeep it under 180 words. Professional but warm.
Knowledge sources
Click Upload files to add reference documents. Your GPT will search these automatically when relevant.Supported formats: PDF, Word (.docx), plain text, CSVNew connectors (2026):
ChatGPT now supports native connections to Google Drive, Slack, and Power BI as knowledge sources — not just uploaded files. Enterprise admins can enable these in workspace settings under Connections.
In the Configure tab, scroll to Capabilities. Toggle on what your GPT needs:
Web search — for GPTs that need current information
Image generation — for creative or visual tasks
Code interpreter — for data analysis, file processing, or calculations
Only enable what the GPT actually needs. Extra capabilities can confuse it.
Model selection
New in 2026 (Enterprise): ChatGPT Team and Enterprise admins can now pin a specific model version to a GPT — ensuring all users get consistent responses regardless of which model ChatGPT defaults to.For most workplace GPTs, the default model works well. Use model pinning when consistency across time is critical (e.g., a compliance GPT where answer drift is a risk).
New in 2026: ChatGPT’s enhanced Memory now works across sessions. When a user enables Memory, ChatGPT remembers key context from past conversations — including preferences, past decisions, and recurring tasks. This makes Custom GPTs even stickier for power users who return to the same GPT regularly.Memory is controlled per-user in Settings → Personalisation → Memory.
Managing what gets remembered
Users can see what ChatGPT has remembered and delete specific memories at any time. As a GPT builder, include a note in your instructions about what the GPT should prioritise remembering — for example: “If the user shares their job title, team size, or preferred format, remember this for future conversations.”
For enterprise users: data used in Custom GPTs and Projects is not used to train the base ChatGPT model. This is confirmed in the OpenAI Enterprise privacy policy. It is safe to reference for organisations with data sensitivity concerns.
If you are on a ChatGPT Team or Enterprise plan, you can publish a GPT to your workspace so all members can find and use it. In the builder, click Save, then set visibility to Anyone in [workspace name].Your GPT will appear in the team’s GPT library. Members do not need to create their own version — they just open it and use it.
Share via link
You can also generate a direct link to your GPT. Click the three-dot menu on your GPT and select Copy link. Anyone with the link and a ChatGPT account can access it (subject to the visibility setting you chose).This is useful for sharing with external collaborators or embedding a GPT link in a handbook or intranet page.
Actions and external integrations
Advanced GPTs can connect to external tools via Actions — essentially API calls that let the GPT pull data from or push data to other systems. Common use cases: triggering a Zapier workflow, pulling CRM data, or submitting a form.Actions require some technical setup. For most workplace GPTs, knowledge files and instructions are sufficient.
