The GPT Builder: Creation Interface

Status as of November 5, 2026: GPTs are in their final weeks. OpenAI is retiring custom GPTs on December 11, 2026 (February 11, 2027, for certain Enterprise workspaces with an exemption). Personal, Free, Go, Plus, and Pro accounts can no longer create or publish new GPTs. In Business, Enterprise, and Edu workspaces, creation depends on settings; it ceased on October 26, 2026, in affected Enterprise workspaces. Existing GPTs remain usable and editable if your plan allows, until retirement; your conversation history with them will remain accessible afterward. The migration tool (Migrate to plugin, found in My GPTs) converts instructions into a skill, the knowledge base into reference files, and connected applications into apps; custom actions are not transferred. This lesson explains how GPTs work to help you maintain them, migrate them, and adapt their principles to a plugin (Lessons 15–17).

What is a custom GPT?

A custom GPT is a tailored version of ChatGPT, configured for a specific task or domain. The distinction from what you have learned so far is important: custom instructions apply to all your conversations, whereas a GPT is a distinct agent with its own rules, knowledge, and capabilities. Switching GPTs is like switching conversation partners.

This is what makes the tool so useful in practice. You can create one GPT to draft your business emails, another to analyze your financial data, and a third to generate marketing visuals—each with its own dedicated context. While your custom instructions had to remain general to cover all your use cases, each GPT can be radically specialized, and therefore much better at its specific task.

Accessing the GPT Builder

When your workspace still allows it, creation happens via Explore GPTs in the sidebar (or at chatgpt.com/gpts), then Create. This is web-only: mobile apps let you use GPTs, not create them. On personal accounts, you can still edit GPTs you previously created, if your plan permits. The builder offers two ways to work, which we will detail below.

Two modes, two philosophies

The conversational builder is the assisted mode. The builder asks you questions in natural language—what do you want this GPT to do? What should it be called? What tone should it adopt?—and converts your answers into configuration. This is the fastest way to get started and a good exercise to clarify your idea. But it has a structural limitation: you do not precisely control what ends up in the final configuration, and for an advanced GPT, this imprecision eventually becomes costly.

The configuration view is the manual mode, the one this course will teach you to master. Here, you control every parameter: the name displayed for your GPT; the description, visible to users if you publish it; the instructions, i.e., the system prompt, the true heart of your GPT (the next lesson is entirely devoted to this); the conversation starters, the suggestions displayed at launch; the knowledge, the files your GPT can consult; the capabilities—web browsing, image generation (GPT Image), code interpreter; and the actions, connections to external APIs that we will cover in Lessons 9–11. The right reflex: start in “Create” to rough things out, then switch to “Configure” to take full control.

Your turn

A minimal GPT, or its design sheet. Let’s design a “Professional French Proofreader” GPT, intentionally minimal. If your workspace still allows creation, build it in the configuration view; otherwise, write the same sheet on paper: it will become the skill for a plugin in Lesson 16. In the configuration view, give it the name Correcteur Pro FR and the description “Corrects and improves your professional French texts.” For the instructions, write: “You are a professional French proofreader. Correct spelling, grammar, and syntax. Propose clearer rephrasing when the text is awkward. Preserve the tone and register of the original. Explain your main corrections.” Add three starters—“Correct this email,” “Improve this paragraph,” “Check this formal letter”—and enable no capabilities: a proofreader needs neither web browsing nor images, and every superfluous capability is wasted context.

Save, then choose sharing: on a personal account, only you have access; in a workspace, you can open it to specific people, the entire workspace, or via a link, depending on settings (Lesson 7). For a first GPT, keep access restricted while you refine it.

The test panel, your best ally

On the right side of the screen, a preview panel lets you test your GPT in real time: every configuration change is reflected there immediately. Use it systematically, and not just with easy cases. Submit a text riddled with errors, a text that is already perfect, an off-topic message—weaknesses in the instructions appear in these situations. The rule is simple: never publish a GPT without testing it with at least five different queries. A GPT that works in your head does not necessarily work in practice.

Two final pitfalls deserve mention. Do not neglect starters: they guide the user on first contact, and without relevant suggestions, most people do not know where to begin. And resist the temptation to enable everything “just in case”: grant your GPT only the capabilities its mission truly requires.

One last design point: a GPT can use connected applications or actions to your own APIs, but not both at once. Also, if you migrate a GPT, its selected model is not carried over to the plugin.

Official sources: GPTs in ChatGPT (OpenAI Help Center: components, retirement, creation by plan) · Creating and editing GPTs (builder, web only) · Custom GPT retirement and migration FAQ

Key takeaways

• A custom GPT is a ChatGPT agent dedicated to a task, distinct from your global custom instructions
• The builder offers a conversational mode (fast, for rough drafts) and a configuration view (precise, for mastery), on the web
• Key elements: name, description, instructions, starters, Knowledge, capabilities, and applications or actions (not both)
• Enable only necessary capabilities and craft your starters carefully
• Retirement on December 11, 2026: no more creation on personal accounts, migration to plugins