Create your own GPT in 2026 — custom GPTs retire
Yes, you can still create your own GPT — but not the way most guides describe. OpenAI has switched off new GPT creation on personal accounts, and plans to retire custom GPTs altogether on 11 December 2026. Below: what applies today, what happens to the GPTs you have already built, and how to get an AI that knows your own documents anyway.
Can you still create your own GPT?
It depends on the account you are on.
- Personal accounts — Free, Go, Plus and Pro: no. Creating and publishing new GPTs is no longer available. GPTs you already built remain usable and editable as long as your plan and permissions allow it.
- Business, Enterprise and Edu: yes, for now. Creation and editing work when workspace settings and permissions allow it. For affected Enterprise workspaces, new GPT creation is planned to end on 26 October 2026.
So for most business owners searching for "create your own GPT", the answer today is: don't build it as a custom GPT. Use Projects on a personal account, plugins on Business or Enterprise, and your own chatbot or AI agent if it needs to live anywhere other than inside ChatGPT. All three are covered below.
What is a custom GPT?
A custom GPT is a version of ChatGPT you have preconfigured with your own instructions and your own documents, so it answers from your situation without you repeating the context every time. You build it without code, directly in ChatGPT.
A typical example: you upload the product manual as a PDF, and support asks the GPT instead of searching the document. It answers from your material rather than from everything it read on the internet.
Three things made them popular:
- Lower risk of hallucinations. Pointing the model at your own documents gives it a clear basis to answer from. The risk does not disappear, but it drops. If the answer should always be one of a few fixed options, such as a category or a yes/no, models like Jev cannot make things up at all.
- No code. Name, description, instructions and files go into a form.
- The same answer every time. A pre-prompted GPT means the whole team gets the same result to the same question.
The abbreviation confuses people, understandably. GPT stands for Generative Pre-trained Transformer and is the name of the model family behind ChatGPT. A custom GPT is therefore not its own model — it is the same model with your instructions and files on top. If you want the technology underneath, we have explained what an LLM is.
Custom GPTs are being retired — the timeline
OpenAI states plainly that it plans to retire custom GPTs and help creators migrate them to plugins. The transition affects all ChatGPT plans.
The communicated dates apply to affected Enterprise workspaces; other plans are expected to follow the same timeline:
- 11 September 2026 — admin notice goes out to organisations.
- 22 September 2026 — target for the migration experience and the user banner.
- 26 October 2026 — creation of new custom GPTs ends. Publish any drafts you want to migrate before then.
- 11 December 2026 — retirement. Custom GPTs stop running and leave the GPT directory.
Dates can change, and OpenAI notes that a target date does not guarantee every account can migrate that day. Follow the notice in your own account. Worth knowing: a public GPT created in an affected Enterprise workspace is included even if you use it from a personal account — what matters is where the GPT was created, not which plan you are on.
What happens to the GPTs you have already built?
They keep working until the retirement date. But some details are worth knowing before you migrate:
- The instructions become a skill in the new plugin, knowledge files become reference files, and connected apps carry over as apps.
- Custom actions do not transfer. If your GPT depends on an API call of your own, that stops working until you rebuild the integration, which may require your own MCP server and technical setup.
- The selected model does not carry over, and Enterprise workspace defaults apply there.
- Sharing settings do not carry over. The migrated plugin starts private, and the people who used your GPT do not automatically get access to the replacement.
- Conversations do not move.
- The original becomes read-only after migration and cannot be deleted. Finish and publish the edits you need first — drafts and unpublished changes do not transfer.
Migration itself runs from My GPTs → Migrate to plugin, once the feature reaches your account. Expect to test the replacement against a few familiar prompts before switching over.
Projects — the closest thing to your own GPT today
If the goal is "a ChatGPT that knows our documents and writes in our tone", Projects is the closest you get today — and it works on every account, including Free.
A project keeps chats, files and custom instructions in one place. Project instructions apply only inside that project and override your global custom instructions, which is effectively the same mechanism a custom GPT relied on. You can also add Google Drive files and Slack channels as project sources by pasting a link.
The file limit per project depends on the plan: 5 files on Free, 25 on Go and Plus, 40 on Pro, Business, Enterprise and Edu. Projects cost nothing extra.
Sharing is available on every plan — Free up to 5 collaborators, Go and Plus up to 10, Pro up to 100. Shared projects automatically use project-only memory, so they cannot reach members' personal memories or custom instructions.
The difference from a custom GPT, in OpenAI's own framing: a GPT was single-player and static — an expert packaged knowledge for others. A shared project is multiplayer and grows over time as everyone contributes files, chats and instructions.
Plugins — OpenAI's official replacement
A plugin packages reusable instructions (skills), connected apps and other capabilities into a workflow. Some plugins contain only skills and need no external connection at all; others link services such as Google Drive or Slack.
This is where OpenAI wants everyone who built a custom GPT to move. The directory is available across ChatGPT plans, but which plugins you can install depends on plan, workspace, role and region — and in a managed workspace the administrator decides through the installation policy.
One thing that is often misread in a business context: a plugin does not grant access to anything the connected account cannot already see. Permissions in Google Drive or Slack still apply, and installation does not bypass them.
Your own chatbot or AI agent instead
Custom GPTs, Projects and plugins share one limitation: they live inside ChatGPT. If the solution needs to sit on your website, answer customers around the clock, or do things in your systems, you need something else.
- Your own chatbot, trained on your material and placed on the site. The customer needs no ChatGPT account, and you control the look, the data and the logs.
- An AI agent that performs tasks — books meetings, updates records, runs several steps in sequence — rather than only answering. We have covered what an AI agent is and how they work.
The rule of thumb: if it should answer from your documents, Projects or a plugin is enough. If it should do something, you need an agent.
How to create your own GPT — step by step
Applies to Business, Enterprise and Edu workspaces with the right permission, until retirement.
- Open Explore GPTs in the ChatGPT sidebar, or go to chatgpt.com/gpts
- Select Create
- Choose how to build: the conversational builder, where you describe what you want, or the configuration view, where you fill in the fields yourself
- Fill in name, description and conversation starters — these are the user-facing fields
- Write the instructions. Use explicit step structure ("When X happens → do Y"), prefer concrete "do X" over long lists of prohibitions, and include examples of acceptable and unacceptable output
- Upload knowledge files — up to 20 files, 512 MB each. Use them for reference material, not for rules; rules belong in the instructions
- Test in Preview before saving
- Select Create and choose how to share it
If you are building something new today, build it as a project or a plugin instead. A GPT you create now still has to be migrated before December.
Which one should you pick?
- ChatGPT that knows your documents, on a personal account → Projects.
- A way of working packaged for the whole team, on Business or Enterprise → a plugin.
- A GPT you already built and want to keep → migrate it to a plugin before 11 December 2026.
- Answering customers on your website → your own chatbot.
- Doing things in your systems → an AI agent.
If you want help working out which one actually solves your problem — and building it — we run workshops and consulting engagements in AI.
Frequently asked questions
Can you create your own GPT in 2026?
Not on personal accounts. Free, Go, Plus and Pro can no longer create or publish new GPTs, though they can still use and edit the ones they already built. On Business, Enterprise and Edu it works for now — for affected Enterprise workspaces, new creation is planned to end on 26 October 2026.
When are custom GPTs retiring?
11 December 2026 under OpenAI's plan. Custom GPTs then stop running and leave the GPT directory. The transition affects all ChatGPT plans, and the dates can change — follow the notice in your own account.
How do I build my own AI assistant?
The simplest route today is a project in ChatGPT: add your documents as files and write project instructions for tone and way of working. It works on every account, including Free, and costs nothing extra. If the assistant needs to sit on your website or act in your systems, you need your own chatbot or an AI agent instead.
What is the difference between a custom GPT and a project?
A GPT was single-player and static — an expert packaged knowledge for others to use. A shared project is multiplayer: members see each other's contributions and the project grows over time through files, chats and instructions. Project instructions override your global custom instructions, just as a GPT's instructions did.
What happens to the GPTs I already built?
They work until retirement and can be migrated to a plugin from My GPTs. Instructions become a skill and knowledge files become reference files, but custom actions, model selection, sharing settings and conversations do not carry over. The original becomes read-only after migration, so finish your edits first.
Does a custom GPT reduce the risk of hallucinations?
It reduces the risk but does not remove it. Limiting the model to your own documents gives it a clear basis to answer from, but it can still misread them or fill gaps on its own.
When should you choose an AI agent instead of a GPT?
A GPT or a project is built for well-defined tasks: answering from documents, keeping a certain tone, repeating a format. If you need something that acts in your systems — books meetings, updates records, runs several steps in sequence — a proper AI agent is the right tool.
Updated: 2026-09-22