Custom GPTs: how to create one, test it and plan its move to plugins

What is a custom GPT and what is it good for at work?

OpenAI describes GPTs, also called custom GPTs, as versions of ChatGPT configured for a specific purpose. A GPT can combine instructions that set behaviour and limits, conversation starters, knowledge files, capabilities such as web search, image generation, canvas and data analysis, and either apps or actions that call external APIs, but not both.

They suit repeated tasks with stable rules, such as checking proposals against a house style or answering questions from a handbook. GPTs do not use saved memory, custom instructions or previous conversations, so every chat starts fresh.

Are custom GPTs being retired?

Yes. OpenAI's help centre says custom GPTs are scheduled to retire on 11 December 2026, or 11 February 2027 for Enterprise workspaces with an approved deferral, and that the change affects all ChatGPT plans. OpenAI recommends moving workflows to plugins, which bring reusable instructions, reference files and connected apps together.

Existing GPTs stay usable until retirement, and past conversations remain accessible afterwards. OpenAI lists 26 October 2026 as the date new GPT creation ends in affected Enterprise workspaces. Dates can change, so follow your account's notice.

Who can still create a custom GPT?

All signed-in ChatGPT users can use GPTs they have access to. Creating and publishing new GPTs is not available on personal accounts, including Free, Go, Plus and Pro. In Business, Enterprise and Edu workspaces, it depends on workspace settings and permissions. Creation happens on the web; the ChatGPT phone apps only let you use GPTs.

If you cannot create one, the design work in this tutorial still applies: you can write the same instructions and reference files into a plugin using Plugin Creator.

How do I create a custom GPT?

In an eligible workspace, open Explore GPTs or go to chatgpt.com/gpts and select Create. Use the conversational builder to describe what you want, or the configuration view to fill in the fields directly: name, description, conversation starters, instructions, knowledge, a recommended model, capabilities and actions.

Knowledge accepts up to 20 files of up to 512 MB each. Test in Preview, then select Create. Edits save to a draft, Update applies them to an existing GPT and version history lets you restore earlier versions.

What is a good 10-minute custom GPT starter exercise?

Use a fictional 18-person software reseller, Quayside Systems, whose proposals keep breaking its own style guide. Write a one-page mock style guide with eight rules.

Minutes 1 to 2: name the GPT Proposal Style Checker and write a one-line description saying who it is for. Minutes 3 to 5: write step-by-step instructions: when the user pastes a draft, check it against each rule, list every breach with the rule number and a suggested fix, then give a verdict. Add one acceptable and one unacceptable example sentence. Minutes 6 to 7: upload the mock style guide as knowledge and add three conversation starters. Minutes 8 to 10: in Preview, paste an invented proposal with three deliberate breaches and check that the GPT finds all three with the right rule numbers.

No create option? Write the same specification as a plugin skill and run the same test.

How do I get better results from a custom GPT?

OpenAI's guidance on instructions and knowledge is the best starting point.

What should I check before sharing a custom GPT or adding actions?

Follow your employer's AI policy on which files may go into knowledge and who a GPT can be shared with. OpenAI says builders cannot see individual conversations users have with their GPTs. Business, Enterprise and Edu data is not used for training by default; on consumer plans, training depends on whether you have opted out.

Actions and apps send relevant parts of a user's input to third-party services. OpenAI states it does not audit or control how those services use or store data, and ChatGPT may ask users to approve a request before it runs. Public GPTs with actions need a privacy policy URL, and workspace admins can limit which action domains are allowed.

How do I migrate a custom GPT to a plugin, and where do GPTs fall short?

When migration is available, go to My GPTs and select Migrate to plugin. OpenAI says instructions become a skill, knowledge files become reference files and connected apps become apps. Migration uses the latest published version, the original becomes read-only, the selected model does not carry over and the plugin starts private. Custom actions do not transfer and need rebuilding, so compare the plugin's answers with familiar prompts before switching.

Beyond retirement, GPTs cannot be embedded in your own website, actions are not available in Pro mode, and GPTs cannot be created on mobile.

How does Tosheo.ai teach custom GPTs?

Tosheo.ai treats custom GPTs as a lesson in specifying a reusable assistant: instructions, reference material, test cases and a migration check. It sits in the AI Agents Development category, which starts with no-code practice and offers optional Code lessons for building agents in code. It is not a bootcamp or a job guarantee. Each 10 to 15 minute lesson ends with a task marked against a visible rubric. Start with the 7-day free trial; no credit card is needed.

Sources and scope

OpenAI Help Center, GPTs in ChatGPT, checked 19 October 2026 ↗

OpenAI Help Center, Creating and editing GPTs, checked 19 October 2026 ↗

OpenAI Help Center, Custom GPT retirement and migration FAQ, checked 19 October 2026 ↗

OpenAI Help Center, Configuring actions in GPTs, checked 19 October 2026 ↗

OpenAI Help Center, Plugins in ChatGPT, checked 19 October 2026 ↗